zstd-ruby 1.4.0.0 → 1.4.9.0

data/ext/zstdruby/libzstd/decompress/zstd_ddict.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2021, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -14,18 +14,18 @@
 /*-*******************************************************
 *  Dependencies
 *********************************************************/
-#include <string.h>      /* memcpy, memmove, memset */
-#include "cpu.h"         /* bmi2 */
-#include "mem.h"         /* low level memory routines */
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
+#include "../common/cpu.h"         /* bmi2 */
+#include "../common/mem.h"         /* low level memory routines */
 #define FSE_STATIC_LINKING_ONLY
-#include "fse.h"
+#include "../common/fse.h"
 #define HUF_STATIC_LINKING_ONLY
-#include "huf.h"
+#include "../common/huf.h"
 #include "zstd_decompress_internal.h"
 #include "zstd_ddict.h"
 
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
-#  include "zstd_legacy.h"
+#  include "../legacy/zstd_legacy.h"
 #endif
 
 
@@ -65,6 +65,10 @@ void ZSTD_copyDDictParameters(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
     dctx->virtualStart = ddict->dictContent;
     dctx->dictEnd = (const BYTE*)ddict->dictContent + ddict->dictSize;
     dctx->previousDstEnd = dctx->dictEnd;
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    dctx->dictContentBeginForFuzzing = dctx->prefixStart;
+    dctx->dictContentEndForFuzzing = dctx->previousDstEnd;
+#endif
     if (ddict->entropyPresent) {
         dctx->litEntropy = 1;
         dctx->fseEntropy = 1;
@@ -107,7 +111,7 @@ ZSTD_loadEntropy_intoDDict(ZSTD_DDict* ddict,
     /* load entropy tables */
     RETURN_ERROR_IF(ZSTD_isError(ZSTD_loadDEntropy(
             &ddict->entropy, ddict->dictContent, ddict->dictSize)),
-        dictionary_corrupted);
+        dictionary_corrupted, "");
     ddict->entropyPresent = 1;
     return 0;
 }
@@ -123,17 +127,17 @@ static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict,
         ddict->dictContent = dict;
         if (!dict) dictSize = 0;
     } else {
-        void* const internalBuffer = ZSTD_malloc(dictSize, ddict->cMem);
+        void* const internalBuffer = ZSTD_customMalloc(dictSize, ddict->cMem);
         ddict->dictBuffer = internalBuffer;
         ddict->dictContent = internalBuffer;
         if (!internalBuffer) return ERROR(memory_allocation);
-        memcpy(internalBuffer, dict, dictSize);
+        ZSTD_memcpy(internalBuffer, dict, dictSize);
     }
     ddict->dictSize = dictSize;
     ddict->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
 
     /* parse dictionary content */
-    FORWARD_IF_ERROR( ZSTD_loadEntropy_intoDDict(ddict, dictContentType) );
+    FORWARD_IF_ERROR( ZSTD_loadEntropy_intoDDict(ddict, dictContentType) , "");
 
     return 0;
 }
@@ -143,9 +147,9 @@ ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
                                       ZSTD_dictContentType_e dictContentType,
                                       ZSTD_customMem customMem)
 {
-    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
 
-    {   ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_malloc(sizeof(ZSTD_DDict), customMem);
+    {   ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_customMalloc(sizeof(ZSTD_DDict), customMem);
         if (ddict == NULL) return NULL;
         ddict->cMem = customMem;
         {   size_t const initResult = ZSTD_initDDict_internal(ddict,
@@ -194,7 +198,7 @@ const ZSTD_DDict* ZSTD_initStaticDDict(
     if ((size_t)sBuffer & 7) return NULL;   /* 8-aligned */
     if (sBufferSize < neededSpace) return NULL;
     if (dictLoadMethod == ZSTD_dlm_byCopy) {
-        memcpy(ddict+1, dict, dictSize);  /* local copy */
+        ZSTD_memcpy(ddict+1, dict, dictSize);  /* local copy */
         dict = ddict+1;
     }
     if (ZSTD_isError( ZSTD_initDDict_internal(ddict,
@@ -209,8 +213,8 @@ size_t ZSTD_freeDDict(ZSTD_DDict* ddict)
 {
     if (ddict==NULL) return 0;   /* support free on NULL */
     {   ZSTD_customMem const cMem = ddict->cMem;
-        ZSTD_free(ddict->dictBuffer, cMem);
-        ZSTD_free(ddict, cMem);
+        ZSTD_customFree(ddict->dictBuffer, cMem);
+        ZSTD_customFree(ddict, cMem);
         return 0;
     }
 }

data/ext/zstdruby/libzstd/decompress/zstd_ddict.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2021, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -15,8 +15,8 @@
 /*-*******************************************************
  *  Dependencies
  *********************************************************/
-#include <stddef.h>   /* size_t */
-#include "zstd.h"     /* ZSTD_DDict, and several public functions */
+#include "../common/zstd_deps.h"   /* size_t */
+#include "../zstd.h"     /* ZSTD_DDict, and several public functions */
 
 
 /*-*******************************************************

data/ext/zstdruby/libzstd/decompress/zstd_decompress.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2021, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -55,23 +55,163 @@
 /*-*******************************************************
 *  Dependencies
 *********************************************************/
-#include <string.h>      /* memcpy, memmove, memset */
-#include "cpu.h"         /* bmi2 */
-#include "mem.h"         /* low level memory routines */
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
+#include "../common/cpu.h"         /* bmi2 */
+#include "../common/mem.h"         /* low level memory routines */
+#include "../common/zstd_trace.h"
 #define FSE_STATIC_LINKING_ONLY
-#include "fse.h"
+#include "../common/fse.h"
 #define HUF_STATIC_LINKING_ONLY
-#include "huf.h"
-#include "zstd_internal.h"  /* blockProperties_t */
+#include "../common/huf.h"
+#include "../common/xxhash.h" /* XXH64_reset, XXH64_update, XXH64_digest, XXH64 */
+#include "../common/zstd_internal.h"  /* blockProperties_t */
 #include "zstd_decompress_internal.h"   /* ZSTD_DCtx */
 #include "zstd_ddict.h"  /* ZSTD_DDictDictContent */
 #include "zstd_decompress_block.h"   /* ZSTD_decompressBlock_internal */
 
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
-#  include "zstd_legacy.h"
+#  include "../legacy/zstd_legacy.h"
 #endif
 
 
+
+/*************************************
+ * Multiple DDicts Hashset internals *
+ *************************************/
+
+#define DDICT_HASHSET_MAX_LOAD_FACTOR_COUNT_MULT 4
+#define DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT 3   /* These two constants represent SIZE_MULT/COUNT_MULT load factor without using a float.
+                                                     * Currently, that means a 0.75 load factor.
+                                                     * So, if count * COUNT_MULT / size * SIZE_MULT != 0, then we've exceeded
+                                                     * the load factor of the ddict hash set.
+                                                     */
+
+#define DDICT_HASHSET_TABLE_BASE_SIZE 64
+#define DDICT_HASHSET_RESIZE_FACTOR 2
+
+/* Hash function to determine starting position of dict insertion within the table
+ * Returns an index between [0, hashSet->ddictPtrTableSize]
+ */
+static size_t ZSTD_DDictHashSet_getIndex(const ZSTD_DDictHashSet* hashSet, U32 dictID) {
+    const U64 hash = XXH64(&dictID, sizeof(U32), 0);
+    /* DDict ptr table size is a multiple of 2, use size - 1 as mask to get index within [0, hashSet->ddictPtrTableSize) */
+    return hash & (hashSet->ddictPtrTableSize - 1);
+}
+
+/* Adds DDict to a hashset without resizing it.
+ * If inserting a DDict with a dictID that already exists in the set, replaces the one in the set.
+ * Returns 0 if successful, or a zstd error code if something went wrong.
+ */
+static size_t ZSTD_DDictHashSet_emplaceDDict(ZSTD_DDictHashSet* hashSet, const ZSTD_DDict* ddict) {
+    const U32 dictID = ZSTD_getDictID_fromDDict(ddict);
+    size_t idx = ZSTD_DDictHashSet_getIndex(hashSet, dictID);
+    const size_t idxRangeMask = hashSet->ddictPtrTableSize - 1;
+    RETURN_ERROR_IF(hashSet->ddictPtrCount == hashSet->ddictPtrTableSize, GENERIC, "Hash set is full!");
+    DEBUGLOG(4, "Hashed index: for dictID: %u is %zu", dictID, idx);
+    while (hashSet->ddictPtrTable[idx] != NULL) {
+        /* Replace existing ddict if inserting ddict with same dictID */
+        if (ZSTD_getDictID_fromDDict(hashSet->ddictPtrTable[idx]) == dictID) {
+            DEBUGLOG(4, "DictID already exists, replacing rather than adding");
+            hashSet->ddictPtrTable[idx] = ddict;
+            return 0;
+        }
+        idx &= idxRangeMask;
+        idx++;
+    }
+    DEBUGLOG(4, "Final idx after probing for dictID %u is: %zu", dictID, idx);
+    hashSet->ddictPtrTable[idx] = ddict;
+    hashSet->ddictPtrCount++;
+    return 0;
+}
+
+/* Expands hash table by factor of DDICT_HASHSET_RESIZE_FACTOR and
+ * rehashes all values, allocates new table, frees old table.
+ * Returns 0 on success, otherwise a zstd error code.
+ */
+static size_t ZSTD_DDictHashSet_expand(ZSTD_DDictHashSet* hashSet, ZSTD_customMem customMem) {
+    size_t newTableSize = hashSet->ddictPtrTableSize * DDICT_HASHSET_RESIZE_FACTOR;
+    const ZSTD_DDict** newTable = (const ZSTD_DDict**)ZSTD_customCalloc(sizeof(ZSTD_DDict*) * newTableSize, customMem);
+    const ZSTD_DDict** oldTable = hashSet->ddictPtrTable;
+    size_t oldTableSize = hashSet->ddictPtrTableSize;
+    size_t i;
+
+    DEBUGLOG(4, "Expanding DDict hash table! Old size: %zu new size: %zu", oldTableSize, newTableSize);
+    RETURN_ERROR_IF(!newTable, memory_allocation, "Expanded hashset allocation failed!");
+    hashSet->ddictPtrTable = newTable;
+    hashSet->ddictPtrTableSize = newTableSize;
+    hashSet->ddictPtrCount = 0;
+    for (i = 0; i < oldTableSize; ++i) {
+        if (oldTable[i] != NULL) {
+            FORWARD_IF_ERROR(ZSTD_DDictHashSet_emplaceDDict(hashSet, oldTable[i]), "");
+        }
+    }
+    ZSTD_customFree((void*)oldTable, customMem);
+    DEBUGLOG(4, "Finished re-hash");
+    return 0;
+}
+
+/* Fetches a DDict with the given dictID
+ * Returns the ZSTD_DDict* with the requested dictID. If it doesn't exist, then returns NULL.
+ */
+static const ZSTD_DDict* ZSTD_DDictHashSet_getDDict(ZSTD_DDictHashSet* hashSet, U32 dictID) {
+    size_t idx = ZSTD_DDictHashSet_getIndex(hashSet, dictID);
+    const size_t idxRangeMask = hashSet->ddictPtrTableSize - 1;
+    DEBUGLOG(4, "Hashed index: for dictID: %u is %zu", dictID, idx);
+    for (;;) {
+        size_t currDictID = ZSTD_getDictID_fromDDict(hashSet->ddictPtrTable[idx]);
+        if (currDictID == dictID || currDictID == 0) {
+            /* currDictID == 0 implies a NULL ddict entry */
+            break;
+        } else {
+            idx &= idxRangeMask;    /* Goes to start of table when we reach the end */
+            idx++;
+        }
+    }
+    DEBUGLOG(4, "Final idx after probing for dictID %u is: %zu", dictID, idx);
+    return hashSet->ddictPtrTable[idx];
+}
+
+/* Allocates space for and returns a ddict hash set
+ * The hash set's ZSTD_DDict* table has all values automatically set to NULL to begin with.
+ * Returns NULL if allocation failed.
+ */
+static ZSTD_DDictHashSet* ZSTD_createDDictHashSet(ZSTD_customMem customMem) {
+    ZSTD_DDictHashSet* ret = (ZSTD_DDictHashSet*)ZSTD_customMalloc(sizeof(ZSTD_DDictHashSet), customMem);
+    DEBUGLOG(4, "Allocating new hash set");
+    ret->ddictPtrTable = (const ZSTD_DDict**)ZSTD_customCalloc(DDICT_HASHSET_TABLE_BASE_SIZE * sizeof(ZSTD_DDict*), customMem);
+    ret->ddictPtrTableSize = DDICT_HASHSET_TABLE_BASE_SIZE;
+    ret->ddictPtrCount = 0;
+    if (!ret || !ret->ddictPtrTable) {
+        return NULL;
+    }
+    return ret;
+}
+
+/* Frees the table of ZSTD_DDict* within a hashset, then frees the hashset itself.
+ * Note: The ZSTD_DDict* within the table are NOT freed.
+ */
+static void ZSTD_freeDDictHashSet(ZSTD_DDictHashSet* hashSet, ZSTD_customMem customMem) {
+    DEBUGLOG(4, "Freeing ddict hash set");
+    if (hashSet && hashSet->ddictPtrTable) {
+        ZSTD_customFree((void*)hashSet->ddictPtrTable, customMem);
+    }
+    if (hashSet) {
+        ZSTD_customFree(hashSet, customMem);
+    }
+}
+
+/* Public function: Adds a DDict into the ZSTD_DDictHashSet, possibly triggering a resize of the hash set.
+ * Returns 0 on success, or a ZSTD error.
+ */
+static size_t ZSTD_DDictHashSet_addDDict(ZSTD_DDictHashSet* hashSet, const ZSTD_DDict* ddict, ZSTD_customMem customMem) {
+    DEBUGLOG(4, "Adding dict ID: %u to hashset with - Count: %zu Tablesize: %zu", ZSTD_getDictID_fromDDict(ddict), hashSet->ddictPtrCount, hashSet->ddictPtrTableSize);
+    if (hashSet->ddictPtrCount * DDICT_HASHSET_MAX_LOAD_FACTOR_COUNT_MULT / hashSet->ddictPtrTableSize * DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT != 0) {
+        FORWARD_IF_ERROR(ZSTD_DDictHashSet_expand(hashSet, customMem), "");
+    }
+    FORWARD_IF_ERROR(ZSTD_DDictHashSet_emplaceDDict(hashSet, ddict), "");
+    return 0;
+}
+
 /*-*************************************************************
 *   Context management
 ***************************************************************/
@@ -88,20 +228,25 @@ size_t ZSTD_estimateDCtxSize(void) { return sizeof(ZSTD_DCtx); }
 
 static size_t ZSTD_startingInputLength(ZSTD_format_e format)
 {
-    size_t const startingInputLength = (format==ZSTD_f_zstd1_magicless) ?
-                    ZSTD_FRAMEHEADERSIZE_PREFIX - ZSTD_FRAMEIDSIZE :
-                    ZSTD_FRAMEHEADERSIZE_PREFIX;
-    ZSTD_STATIC_ASSERT(ZSTD_FRAMEHEADERSIZE_PREFIX >= ZSTD_FRAMEIDSIZE);
+    size_t const startingInputLength = ZSTD_FRAMEHEADERSIZE_PREFIX(format);
     /* only supports formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless */
     assert( (format == ZSTD_f_zstd1) || (format == ZSTD_f_zstd1_magicless) );
     return startingInputLength;
 }
 
+static void ZSTD_DCtx_resetParameters(ZSTD_DCtx* dctx)
+{
+    assert(dctx->streamStage == zdss_init);
+    dctx->format = ZSTD_f_zstd1;
+    dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
+    dctx->outBufferMode = ZSTD_bm_buffered;
+    dctx->forceIgnoreChecksum = ZSTD_d_validateChecksum;
+    dctx->refMultipleDDicts = ZSTD_rmd_refSingleDDict;
+}
+
 static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
 {
-    dctx->format = ZSTD_f_zstd1;  /* ZSTD_decompressBegin() invokes ZSTD_startingInputLength() with argument dctx->format */
     dctx->staticSize  = 0;
-    dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
     dctx->ddict       = NULL;
     dctx->ddictLocal  = NULL;
     dctx->dictEnd     = NULL;
@@ -114,7 +259,13 @@ static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
     dctx->legacyContext = NULL;
     dctx->previousLegacyVersion = 0;
     dctx->noForwardProgress = 0;
+    dctx->oversizedDuration = 0;
     dctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+    dctx->ddictSet = NULL;
+    ZSTD_DCtx_resetParameters(dctx);
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    dctx->dictContentEndForFuzzing = NULL;
+#endif
 }
 
 ZSTD_DCtx* ZSTD_initStaticDCtx(void *workspace, size_t workspaceSize)
@@ -132,9 +283,9 @@ ZSTD_DCtx* ZSTD_initStaticDCtx(void *workspace, size_t workspaceSize)
 
 ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
 {
-    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
 
-    {   ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_malloc(sizeof(*dctx), customMem);
+    {   ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_customMalloc(sizeof(*dctx), customMem);
         if (!dctx) return NULL;
         dctx->customMem = customMem;
         ZSTD_initDCtx_internal(dctx);
@@ -162,13 +313,17 @@ size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
     RETURN_ERROR_IF(dctx->staticSize, memory_allocation, "not compatible with static DCtx");
     {   ZSTD_customMem const cMem = dctx->customMem;
         ZSTD_clearDict(dctx);
-        ZSTD_free(dctx->inBuff, cMem);
+        ZSTD_customFree(dctx->inBuff, cMem);
         dctx->inBuff = NULL;
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
         if (dctx->legacyContext)
             ZSTD_freeLegacyStreamContext(dctx->legacyContext, dctx->previousLegacyVersion);
 #endif
-        ZSTD_free(dctx, cMem);
+        if (dctx->ddictSet) {
+            ZSTD_freeDDictHashSet(dctx->ddictSet, cMem);
+            dctx->ddictSet = NULL;
+        }
+        ZSTD_customFree(dctx, cMem);
         return 0;
     }
 }
@@ -177,7 +332,30 @@ size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
 void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
 {
     size_t const toCopy = (size_t)((char*)(&dstDCtx->inBuff) - (char*)dstDCtx);
-    memcpy(dstDCtx, srcDCtx, toCopy);  /* no need to copy workspace */
+    ZSTD_memcpy(dstDCtx, srcDCtx, toCopy);  /* no need to copy workspace */
+}
+
+/* Given a dctx with a digested frame params, re-selects the correct ZSTD_DDict based on
+ * the requested dict ID from the frame. If there exists a reference to the correct ZSTD_DDict, then
+ * accordingly sets the ddict to be used to decompress the frame.
+ *
+ * If no DDict is found, then no action is taken, and the ZSTD_DCtx::ddict remains as-is.
+ *
+ * ZSTD_d_refMultipleDDicts must be enabled for this function to be called.
+ */
+static void ZSTD_DCtx_selectFrameDDict(ZSTD_DCtx* dctx) {
+    assert(dctx->refMultipleDDicts && dctx->ddictSet);
+    DEBUGLOG(4, "Adjusting DDict based on requested dict ID from frame");
+    if (dctx->ddict) {
+        const ZSTD_DDict* frameDDict = ZSTD_DDictHashSet_getDDict(dctx->ddictSet, dctx->fParams.dictID);
+        if (frameDDict) {
+            DEBUGLOG(4, "DDict found!");
+            ZSTD_clearDict(dctx);
+            dctx->dictID = dctx->fParams.dictID;
+            dctx->ddict = frameDDict;
+            dctx->dictUses = ZSTD_use_indefinitely;
+        }
+    }
 }
 
 
@@ -211,7 +389,7 @@ unsigned ZSTD_isFrame(const void* buffer, size_t size)
 static size_t ZSTD_frameHeaderSize_internal(const void* src, size_t srcSize, ZSTD_format_e format)
 {
     size_t const minInputSize = ZSTD_startingInputLength(format);
-    RETURN_ERROR_IF(srcSize < minInputSize, srcSize_wrong);
+    RETURN_ERROR_IF(srcSize < minInputSize, srcSize_wrong, "");
 
     {   BYTE const fhd = ((const BYTE*)src)[minInputSize-1];
         U32 const dictID= fhd & 3;
@@ -244,7 +422,7 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
     const BYTE* ip = (const BYTE*)src;
     size_t const minInputSize = ZSTD_startingInputLength(format);
 
-    memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzer do not understand that zfhPtr is only going to be read only if return value is zero, since they are 2 different signals */
+    ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzer do not understand that zfhPtr is only going to be read only if return value is zero, since they are 2 different signals */
     if (srcSize < minInputSize) return minInputSize;
     RETURN_ERROR_IF(src==NULL, GENERIC, "invalid parameter");
 
@@ -254,12 +432,12 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
             /* skippable frame */
             if (srcSize < ZSTD_SKIPPABLEHEADERSIZE)
                 return ZSTD_SKIPPABLEHEADERSIZE; /* magic number + frame length */
-            memset(zfhPtr, 0, sizeof(*zfhPtr));
+            ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));
             zfhPtr->frameContentSize = MEM_readLE32((const char *)src + ZSTD_FRAMEIDSIZE);
             zfhPtr->frameType = ZSTD_skippableFrame;
             return 0;
         }
-        RETURN_ERROR(prefix_unknown);
+        RETURN_ERROR(prefix_unknown, "");
     }
 
     /* ensure there is enough `srcSize` to fully read/decode frame header */
@@ -283,7 +461,7 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
         if (!singleSegment) {
             BYTE const wlByte = ip[pos++];
             U32 const windowLog = (wlByte >> 3) + ZSTD_WINDOWLOG_ABSOLUTEMIN;
-            RETURN_ERROR_IF(windowLog > ZSTD_WINDOWLOG_MAX, frameParameter_windowTooLarge);
+            RETURN_ERROR_IF(windowLog > ZSTD_WINDOWLOG_MAX, frameParameter_windowTooLarge, "");
             windowSize = (1ULL << windowLog);
             windowSize += (windowSize >> 3) * (wlByte&7);
         }
@@ -355,13 +533,16 @@ static size_t readSkippableFrameSize(void const* src, size_t srcSize)
     size_t const skippableHeaderSize = ZSTD_SKIPPABLEHEADERSIZE;
     U32 sizeU32;
 
-    RETURN_ERROR_IF(srcSize < ZSTD_SKIPPABLEHEADERSIZE, srcSize_wrong);
+    RETURN_ERROR_IF(srcSize < ZSTD_SKIPPABLEHEADERSIZE, srcSize_wrong, "");
 
     sizeU32 = MEM_readLE32((BYTE const*)src + ZSTD_FRAMEIDSIZE);
     RETURN_ERROR_IF((U32)(sizeU32 + ZSTD_SKIPPABLEHEADERSIZE) < sizeU32,
-                    frameParameter_unsupported);
-
-    return skippableHeaderSize + sizeU32;
+                    frameParameter_unsupported, "");
+    {
+        size_t const skippableSize = skippableHeaderSize + sizeU32;
+        RETURN_ERROR_IF(skippableSize > srcSize, srcSize_wrong, "");
+        return skippableSize;
+    }
 }
 
 /** ZSTD_findDecompressedSize() :
@@ -373,16 +554,15 @@ unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
 {
     unsigned long long totalDstSize = 0;
 
-    while (srcSize >= ZSTD_FRAMEHEADERSIZE_PREFIX) {
+    while (srcSize >= ZSTD_startingInputLength(ZSTD_f_zstd1)) {
         U32 const magicNumber = MEM_readLE32(src);
 
         if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
             size_t const skippableSize = readSkippableFrameSize(src, srcSize);
-            if (ZSTD_isError(skippableSize))
-                return skippableSize;
-            if (srcSize < skippableSize) {
+            if (ZSTD_isError(skippableSize)) {
                 return ZSTD_CONTENTSIZE_ERROR;
             }
+            assert(skippableSize <= srcSize);
 
             src = (const BYTE *)src + skippableSize;
             srcSize -= skippableSize;
@@ -429,20 +609,29 @@ unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize)
 
 /** ZSTD_decodeFrameHeader() :
  * `headerSize` must be the size provided by ZSTD_frameHeaderSize().
+ * If multiple DDict references are enabled, also will choose the correct DDict to use.
  * @return : 0 if success, or an error code, which can be tested using ZSTD_isError() */
 static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t headerSize)
 {
     size_t const result = ZSTD_getFrameHeader_advanced(&(dctx->fParams), src, headerSize, dctx->format);
     if (ZSTD_isError(result)) return result;    /* invalid header */
     RETURN_ERROR_IF(result>0, srcSize_wrong, "headerSize too small");
+
+    /* Reference DDict requested by frame if dctx references multiple ddicts */
+    if (dctx->refMultipleDDicts == ZSTD_rmd_refMultipleDDicts && dctx->ddictSet) {
+        ZSTD_DCtx_selectFrameDDict(dctx);
+    }
+
 #ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
     /* Skip the dictID check in fuzzing mode, because it makes the search
      * harder.
      */
     RETURN_ERROR_IF(dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID),
-                    dictionary_wrong);
+                    dictionary_wrong, "");
 #endif
-    if (dctx->fParams.checksumFlag) XXH64_reset(&dctx->xxhState, 0);
+    dctx->validateChecksum = (dctx->fParams.checksumFlag && !dctx->forceIgnoreChecksum) ? 1 : 0;
+    if (dctx->validateChecksum) XXH64_reset(&dctx->xxhState, 0);
+    dctx->processedCSize += headerSize;
     return 0;
 }
 
@@ -457,7 +646,7 @@ static ZSTD_frameSizeInfo ZSTD_errorFrameSizeInfo(size_t ret)
 static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize)
 {
     ZSTD_frameSizeInfo frameSizeInfo;
-    memset(&frameSizeInfo, 0, sizeof(ZSTD_frameSizeInfo));
+    ZSTD_memset(&frameSizeInfo, 0, sizeof(ZSTD_frameSizeInfo));
 
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
     if (ZSTD_isLegacy(src, srcSize))
@@ -467,6 +656,8 @@ static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize
     if ((srcSize >= ZSTD_SKIPPABLEHEADERSIZE)
         && (MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
         frameSizeInfo.compressedSize = readSkippableFrameSize(src, srcSize);
+        assert(ZSTD_isError(frameSizeInfo.compressedSize) ||
+               frameSizeInfo.compressedSize <= srcSize);
         return frameSizeInfo;
     } else {
         const BYTE* ip = (const BYTE*)src;
@@ -510,7 +701,7 @@ static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize
             ip += 4;
         }
 
-        frameSizeInfo.compressedSize = ip - ipstart;
+        frameSizeInfo.compressedSize = (size_t)(ip - ipstart);
         frameSizeInfo.decompressedBound = (zfh.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN)
                                         ? zfh.frameContentSize
                                         : nbBlocks * zfh.blockSizeMax;
@@ -529,7 +720,6 @@ size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
     return frameSizeInfo.compressedSize;
 }
 
-
 /** ZSTD_decompressBound() :
  *  compatible with legacy mode
  *  `src` must point to the start of a ZSTD frame or a skippeable frame
@@ -546,6 +736,7 @@ unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize)
         unsigned long long const decompressedBound = frameSizeInfo.decompressedBound;
         if (ZSTD_isError(compressedSize) || decompressedBound == ZSTD_CONTENTSIZE_ERROR)
             return ZSTD_CONTENTSIZE_ERROR;
+        assert(srcSize >= compressedSize);
         src = (const BYTE*)src + compressedSize;
         srcSize -= compressedSize;
         bound += decompressedBound;
@@ -558,22 +749,12 @@ unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize)
  *   Frame decoding
  ***************************************************************/
 
-
-void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst)
-{
-    if (dst != dctx->previousDstEnd) {   /* not contiguous */
-        dctx->dictEnd = dctx->previousDstEnd;
-        dctx->virtualStart = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->prefixStart));
-        dctx->prefixStart = dst;
-        dctx->previousDstEnd = dst;
-    }
-}
-
 /** ZSTD_insertBlock() :
-    insert `src` block into `dctx` history. Useful to track uncompressed blocks. */
+ *  insert `src` block into `dctx` history. Useful to track uncompressed blocks. */
 size_t ZSTD_insertBlock(ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize)
 {
-    ZSTD_checkContinuity(dctx, blockStart);
+    DEBUGLOG(5, "ZSTD_insertBlock: %u bytes", (unsigned)blockSize);
+    ZSTD_checkContinuity(dctx, blockStart, blockSize);
     dctx->previousDstEnd = (const char*)blockStart + blockSize;
     return blockSize;
 }
@@ -583,12 +764,12 @@ static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
                           const void* src, size_t srcSize)
 {
     DEBUGLOG(5, "ZSTD_copyRawBlock");
+    RETURN_ERROR_IF(srcSize > dstCapacity, dstSize_tooSmall, "");
     if (dst == NULL) {
         if (srcSize == 0) return 0;
-        RETURN_ERROR(dstBuffer_null);
+        RETURN_ERROR(dstBuffer_null, "");
     }
-    RETURN_ERROR_IF(srcSize > dstCapacity, dstSize_tooSmall);
-    memcpy(dst, src, srcSize);
+    ZSTD_memcpy(dst, src, srcSize);
     return srcSize;
 }
 
@@ -596,15 +777,41 @@ static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity,
                                BYTE b,
                                size_t regenSize)
 {
+    RETURN_ERROR_IF(regenSize > dstCapacity, dstSize_tooSmall, "");
     if (dst == NULL) {
         if (regenSize == 0) return 0;
-        RETURN_ERROR(dstBuffer_null);
+        RETURN_ERROR(dstBuffer_null, "");
     }
-    RETURN_ERROR_IF(regenSize > dstCapacity, dstSize_tooSmall);
-    memset(dst, b, regenSize);
+    ZSTD_memset(dst, b, regenSize);
     return regenSize;
 }
 
+static void ZSTD_DCtx_trace_end(ZSTD_DCtx const* dctx, U64 uncompressedSize, U64 compressedSize, unsigned streaming)
+{
+#if ZSTD_TRACE
+    if (dctx->traceCtx) {
+        ZSTD_Trace trace;
+        ZSTD_memset(&trace, 0, sizeof(trace));
+        trace.version = ZSTD_VERSION_NUMBER;
+        trace.streaming = streaming;
+        if (dctx->ddict) {
+            trace.dictionaryID = ZSTD_getDictID_fromDDict(dctx->ddict);
+            trace.dictionarySize = ZSTD_DDict_dictSize(dctx->ddict);
+            trace.dictionaryIsCold = dctx->ddictIsCold;
+        }
+        trace.uncompressedSize = (size_t)uncompressedSize;
+        trace.compressedSize = (size_t)compressedSize;
+        trace.dctx = dctx;
+        ZSTD_trace_decompress_end(dctx->traceCtx, &trace);
+    }
+#else
+    (void)dctx;
+    (void)uncompressedSize;
+    (void)compressedSize;
+    (void)streaming;
+#endif
+}
+
 
 /*! ZSTD_decompressFrame() :
  * @dctx must be properly initialized
@@ -614,9 +821,10 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
                                    void* dst, size_t dstCapacity,
                              const void** srcPtr, size_t *srcSizePtr)
 {
-    const BYTE* ip = (const BYTE*)(*srcPtr);
-    BYTE* const ostart = (BYTE* const)dst;
-    BYTE* const oend = ostart + dstCapacity;
+    const BYTE* const istart = (const BYTE*)(*srcPtr);
+    const BYTE* ip = istart;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = dstCapacity != 0 ? ostart + dstCapacity : ostart;
     BYTE* op = ostart;
     size_t remainingSrcSize = *srcSizePtr;
 
@@ -624,15 +832,16 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
 
     /* check */
     RETURN_ERROR_IF(
-        remainingSrcSize < ZSTD_FRAMEHEADERSIZE_MIN+ZSTD_blockHeaderSize,
-        srcSize_wrong);
+        remainingSrcSize < ZSTD_FRAMEHEADERSIZE_MIN(dctx->format)+ZSTD_blockHeaderSize,
+        srcSize_wrong, "");
 
     /* Frame Header */
-    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize(ip, ZSTD_FRAMEHEADERSIZE_PREFIX);
+    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize_internal(
+                ip, ZSTD_FRAMEHEADERSIZE_PREFIX(dctx->format), dctx->format);
         if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
         RETURN_ERROR_IF(remainingSrcSize < frameHeaderSize+ZSTD_blockHeaderSize,
-                        srcSize_wrong);
-        FORWARD_IF_ERROR( ZSTD_decodeFrameHeader(dctx, ip, frameHeaderSize) );
+                        srcSize_wrong, "");
+        FORWARD_IF_ERROR( ZSTD_decodeFrameHeader(dctx, ip, frameHeaderSize) , "");
         ip += frameHeaderSize; remainingSrcSize -= frameHeaderSize;
     }
 
@@ -645,28 +854,30 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
 
         ip += ZSTD_blockHeaderSize;
         remainingSrcSize -= ZSTD_blockHeaderSize;
-        RETURN_ERROR_IF(cBlockSize > remainingSrcSize, srcSize_wrong);
+        RETURN_ERROR_IF(cBlockSize > remainingSrcSize, srcSize_wrong, "");
 
         switch(blockProperties.blockType)
         {
         case bt_compressed:
-            decodedSize = ZSTD_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize, /* frame */ 1);
+            decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oend-op), ip, cBlockSize, /* frame */ 1);
             break;
         case bt_raw :
-            decodedSize = ZSTD_copyRawBlock(op, oend-op, ip, cBlockSize);
+            decodedSize = ZSTD_copyRawBlock(op, (size_t)(oend-op), ip, cBlockSize);
             break;
         case bt_rle :
-            decodedSize = ZSTD_setRleBlock(op, oend-op, *ip, blockProperties.origSize);
+            decodedSize = ZSTD_setRleBlock(op, (size_t)(oend-op), *ip, blockProperties.origSize);
             break;
         case bt_reserved :
         default:
-            RETURN_ERROR(corruption_detected);
+            RETURN_ERROR(corruption_detected, "invalid block type");
         }
 
         if (ZSTD_isError(decodedSize)) return decodedSize;
-        if (dctx->fParams.checksumFlag)
+        if (dctx->validateChecksum)
             XXH64_update(&dctx->xxhState, op, decodedSize);
-        op += decodedSize;
+        if (decodedSize != 0)
+            op += decodedSize;
+        assert(ip != NULL);
         ip += cBlockSize;
         remainingSrcSize -= cBlockSize;
         if (blockProperties.lastBlock) break;
@@ -674,22 +885,24 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
 
     if (dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN) {
         RETURN_ERROR_IF((U64)(op-ostart) != dctx->fParams.frameContentSize,
-                        corruption_detected);
+                        corruption_detected, "");
     }
     if (dctx->fParams.checksumFlag) { /* Frame content checksum verification */
-        U32 const checkCalc = (U32)XXH64_digest(&dctx->xxhState);
-        U32 checkRead;
-        RETURN_ERROR_IF(remainingSrcSize<4, checksum_wrong);
-        checkRead = MEM_readLE32(ip);
-        RETURN_ERROR_IF(checkRead != checkCalc, checksum_wrong);
+        RETURN_ERROR_IF(remainingSrcSize<4, checksum_wrong, "");
+        if (!dctx->forceIgnoreChecksum) {
+            U32 const checkCalc = (U32)XXH64_digest(&dctx->xxhState);
+            U32 checkRead;
+            checkRead = MEM_readLE32(ip);
+            RETURN_ERROR_IF(checkRead != checkCalc, checksum_wrong, "");
+        }
         ip += 4;
         remainingSrcSize -= 4;
     }
-
+    ZSTD_DCtx_trace_end(dctx, (U64)(op-ostart), (U64)(ip-istart), /* streaming */ 0);
     /* Allow caller to get size read */
     *srcPtr = ip;
     *srcSizePtr = remainingSrcSize;
-    return op-ostart;
+    return (size_t)(op-ostart);
 }
 
 static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
@@ -709,7 +922,7 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
         dictSize = ZSTD_DDict_dictSize(ddict);
     }
 
-    while (srcSize >= ZSTD_FRAMEHEADERSIZE_PREFIX) {
+    while (srcSize >= ZSTD_startingInputLength(dctx->format)) {
 
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
         if (ZSTD_isLegacy(src, srcSize)) {
@@ -722,7 +935,7 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
             decodedSize = ZSTD_decompressLegacy(dst, dstCapacity, src, frameSize, dict, dictSize);
             if (ZSTD_isError(decodedSize)) return decodedSize;
 
-            assert(decodedSize <=- dstCapacity);
+            assert(decodedSize <= dstCapacity);
             dst = (BYTE*)dst + decodedSize;
             dstCapacity -= decodedSize;
 
@@ -738,9 +951,8 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
                         (unsigned)magicNumber, ZSTD_MAGICNUMBER);
             if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
                 size_t const skippableSize = readSkippableFrameSize(src, srcSize);
-                if (ZSTD_isError(skippableSize))
-                    return skippableSize;
-                RETURN_ERROR_IF(srcSize < skippableSize, srcSize_wrong);
+                FORWARD_IF_ERROR(skippableSize, "readSkippableFrameSize failed");
+                assert(skippableSize <= srcSize);
 
                 src = (const BYTE *)src + skippableSize;
                 srcSize -= skippableSize;
@@ -749,13 +961,13 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
 
         if (ddict) {
             /* we were called from ZSTD_decompress_usingDDict */
-            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDDict(dctx, ddict));
+            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDDict(dctx, ddict), "");
         } else {
             /* this will initialize correctly with no dict if dict == NULL, so
              * use this in all cases but ddict */
-            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDict(dctx, dict, dictSize));
+            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDict(dctx, dict, dictSize), "");
         }
-        ZSTD_checkContinuity(dctx, dst);
+        ZSTD_checkContinuity(dctx, dst, dstCapacity);
 
         {   const size_t res = ZSTD_decompressFrame(dctx, dst, dstCapacity,
                                                     &src, &srcSize);
@@ -763,18 +975,17 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
                 (ZSTD_getErrorCode(res) == ZSTD_error_prefix_unknown)
              && (moreThan1Frame==1),
                 srcSize_wrong,
-                "at least one frame successfully completed, but following "
-                "bytes are garbage: it's more likely to be a srcSize error, "
-                "specifying more bytes than compressed size of frame(s). This "
-                "error message replaces ERROR(prefix_unknown), which would be "
-                "confusing, as the first header is actually correct. Note that "
-                "one could be unlucky, it might be a corruption error instead, "
-                "happening right at the place where we expect zstd magic "
-                "bytes. But this is _much_ less likely than a srcSize field "
-                "error.");
+                "At least one frame successfully completed, "
+                "but following bytes are garbage: "
+                "it's more likely to be a srcSize error, "
+                "specifying more input bytes than size of frame(s). "
+                "Note: one could be unlucky, it might be a corruption error instead, "
+                "happening right at the place where we expect zstd magic bytes. "
+                "But this is _much_ less likely than a srcSize field error.");
             if (ZSTD_isError(res)) return res;
             assert(res <= dstCapacity);
-            dst = (BYTE*)dst + res;
+            if (res != 0)
+                dst = (BYTE*)dst + res;
             dstCapacity -= res;
         }
         moreThan1Frame = 1;
@@ -782,7 +993,7 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
 
     RETURN_ERROR_IF(srcSize, srcSize_wrong, "input not entirely consumed");
 
-    return (BYTE*)dst - (BYTE*)dststart;
+    return (size_t)((BYTE*)dst - (BYTE*)dststart);
 }
 
 size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
@@ -822,7 +1033,7 @@ size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t sr
 #if defined(ZSTD_HEAPMODE) && (ZSTD_HEAPMODE>=1)
     size_t regenSize;
     ZSTD_DCtx* const dctx = ZSTD_createDCtx();
-    RETURN_ERROR_IF(dctx==NULL, memory_allocation);
+    RETURN_ERROR_IF(dctx==NULL, memory_allocation, "NULL pointer!");
     regenSize = ZSTD_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
     ZSTD_freeDCtx(dctx);
     return regenSize;
@@ -840,6 +1051,24 @@ size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t sr
 ****************************************/
 size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx) { return dctx->expected; }
 
+/**
+ * Similar to ZSTD_nextSrcSizeToDecompress(), but when when a block input can be streamed,
+ * we allow taking a partial block as the input. Currently only raw uncompressed blocks can
+ * be streamed.
+ *
+ * For blocks that can be streamed, this allows us to reduce the latency until we produce
+ * output, and avoid copying the input.
+ *
+ * @param inputSize - The total amount of input that the caller currently has.
+ */
+static size_t ZSTD_nextSrcSizeToDecompressWithInputSize(ZSTD_DCtx* dctx, size_t inputSize) {
+    if (!(dctx->stage == ZSTDds_decompressBlock || dctx->stage == ZSTDds_decompressLastBlock))
+        return dctx->expected;
+    if (dctx->bType != bt_raw)
+        return dctx->expected;
+    return MIN(MAX(inputSize, 1), dctx->expected);
+}
+
 ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx) {
     switch(dctx->stage)
     {
@@ -872,8 +1101,10 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
 {
     DEBUGLOG(5, "ZSTD_decompressContinue (srcSize:%u)", (unsigned)srcSize);
     /* Sanity check */
-    RETURN_ERROR_IF(srcSize != dctx->expected, srcSize_wrong, "not allowed");
-    if (dstCapacity) ZSTD_checkContinuity(dctx, dst);
+    RETURN_ERROR_IF(srcSize != ZSTD_nextSrcSizeToDecompressWithInputSize(dctx, srcSize), srcSize_wrong, "not allowed");
+    ZSTD_checkContinuity(dctx, dst, dstCapacity);
+
+    dctx->processedCSize += srcSize;
 
     switch (dctx->stage)
     {
@@ -882,22 +1113,22 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
         if (dctx->format == ZSTD_f_zstd1) {  /* allows header */
             assert(srcSize >= ZSTD_FRAMEIDSIZE);  /* to read skippable magic number */
             if ((MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {        /* skippable frame */
-                memcpy(dctx->headerBuffer, src, srcSize);
+                ZSTD_memcpy(dctx->headerBuffer, src, srcSize);
                 dctx->expected = ZSTD_SKIPPABLEHEADERSIZE - srcSize;  /* remaining to load to get full skippable frame header */
                 dctx->stage = ZSTDds_decodeSkippableHeader;
                 return 0;
         }   }
         dctx->headerSize = ZSTD_frameHeaderSize_internal(src, srcSize, dctx->format);
         if (ZSTD_isError(dctx->headerSize)) return dctx->headerSize;
-        memcpy(dctx->headerBuffer, src, srcSize);
+        ZSTD_memcpy(dctx->headerBuffer, src, srcSize);
         dctx->expected = dctx->headerSize - srcSize;
         dctx->stage = ZSTDds_decodeFrameHeader;
         return 0;
 
     case ZSTDds_decodeFrameHeader:
         assert(src != NULL);
-        memcpy(dctx->headerBuffer + (dctx->headerSize - srcSize), src, srcSize);
-        FORWARD_IF_ERROR(ZSTD_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize));
+        ZSTD_memcpy(dctx->headerBuffer + (dctx->headerSize - srcSize), src, srcSize);
+        FORWARD_IF_ERROR(ZSTD_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize), "");
         dctx->expected = ZSTD_blockHeaderSize;
         dctx->stage = ZSTDds_decodeBlockHeader;
         return 0;
@@ -906,6 +1137,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
         {   blockProperties_t bp;
             size_t const cBlockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
             if (ZSTD_isError(cBlockSize)) return cBlockSize;
+            RETURN_ERROR_IF(cBlockSize > dctx->fParams.blockSizeMax, corruption_detected, "Block Size Exceeds Maximum");
             dctx->expected = cBlockSize;
             dctx->bType = bp.blockType;
             dctx->rleSize = bp.origSize;
@@ -938,49 +1170,66 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
             case bt_compressed:
                 DEBUGLOG(5, "ZSTD_decompressContinue: case bt_compressed");
                 rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 1);
+                dctx->expected = 0;  /* Streaming not supported */
                 break;
             case bt_raw :
+                assert(srcSize <= dctx->expected);
                 rSize = ZSTD_copyRawBlock(dst, dstCapacity, src, srcSize);
+                FORWARD_IF_ERROR(rSize, "ZSTD_copyRawBlock failed");
+                assert(rSize == srcSize);
+                dctx->expected -= rSize;
                 break;
             case bt_rle :
                 rSize = ZSTD_setRleBlock(dst, dstCapacity, *(const BYTE*)src, dctx->rleSize);
+                dctx->expected = 0;  /* Streaming not supported */
                 break;
             case bt_reserved :   /* should never happen */
             default:
-                RETURN_ERROR(corruption_detected);
+                RETURN_ERROR(corruption_detected, "invalid block type");
             }
-            if (ZSTD_isError(rSize)) return rSize;
+            FORWARD_IF_ERROR(rSize, "");
+            RETURN_ERROR_IF(rSize > dctx->fParams.blockSizeMax, corruption_detected, "Decompressed Block Size Exceeds Maximum");
             DEBUGLOG(5, "ZSTD_decompressContinue: decoded size from block : %u", (unsigned)rSize);
             dctx->decodedSize += rSize;
-            if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, dst, rSize);
+            if (dctx->validateChecksum) XXH64_update(&dctx->xxhState, dst, rSize);
+            dctx->previousDstEnd = (char*)dst + rSize;
+
+            /* Stay on the same stage until we are finished streaming the block. */
+            if (dctx->expected > 0) {
+                return rSize;
+            }
 
             if (dctx->stage == ZSTDds_decompressLastBlock) {   /* end of frame */
                 DEBUGLOG(4, "ZSTD_decompressContinue: decoded size from frame : %u", (unsigned)dctx->decodedSize);
                 RETURN_ERROR_IF(
                     dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
                  && dctx->decodedSize != dctx->fParams.frameContentSize,
-                    corruption_detected);
+                    corruption_detected, "");
                 if (dctx->fParams.checksumFlag) {  /* another round for frame checksum */
                     dctx->expected = 4;
                     dctx->stage = ZSTDds_checkChecksum;
                 } else {
+                    ZSTD_DCtx_trace_end(dctx, dctx->decodedSize, dctx->processedCSize, /* streaming */ 1);
                     dctx->expected = 0;   /* ends here */
                     dctx->stage = ZSTDds_getFrameHeaderSize;
                 }
             } else {
                 dctx->stage = ZSTDds_decodeBlockHeader;
                 dctx->expected = ZSTD_blockHeaderSize;
-                dctx->previousDstEnd = (char*)dst + rSize;
             }
             return rSize;
         }
 
     case ZSTDds_checkChecksum:
         assert(srcSize == 4);  /* guaranteed by dctx->expected */
-        {   U32 const h32 = (U32)XXH64_digest(&dctx->xxhState);
-            U32 const check32 = MEM_readLE32(src);
-            DEBUGLOG(4, "ZSTD_decompressContinue: checksum : calculated %08X :: %08X read", (unsigned)h32, (unsigned)check32);
-            RETURN_ERROR_IF(check32 != h32, checksum_wrong);
+        {
+            if (dctx->validateChecksum) {
+                U32 const h32 = (U32)XXH64_digest(&dctx->xxhState);
+                U32 const check32 = MEM_readLE32(src);
+                DEBUGLOG(4, "ZSTD_decompressContinue: checksum : calculated %08X :: %08X read", (unsigned)h32, (unsigned)check32);
+                RETURN_ERROR_IF(check32 != h32, checksum_wrong, "");
+            }
+            ZSTD_DCtx_trace_end(dctx, dctx->decodedSize, dctx->processedCSize, /* streaming */ 1);
             dctx->expected = 0;
             dctx->stage = ZSTDds_getFrameHeaderSize;
             return 0;
@@ -989,7 +1238,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
     case ZSTDds_decodeSkippableHeader:
         assert(src != NULL);
         assert(srcSize <= ZSTD_SKIPPABLEHEADERSIZE);
-        memcpy(dctx->headerBuffer + (ZSTD_SKIPPABLEHEADERSIZE - srcSize), src, srcSize);   /* complete skippable header */
+        ZSTD_memcpy(dctx->headerBuffer + (ZSTD_SKIPPABLEHEADERSIZE - srcSize), src, srcSize);   /* complete skippable header */
         dctx->expected = MEM_readLE32(dctx->headerBuffer + ZSTD_FRAMEIDSIZE);   /* note : dctx->expected can grow seriously large, beyond local buffer size */
         dctx->stage = ZSTDds_skipFrame;
         return 0;
@@ -1001,7 +1250,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
 
     default:
         assert(0);   /* impossible */
-        RETURN_ERROR(GENERIC);   /* some compiler require default to do something */
+        RETURN_ERROR(GENERIC, "impossible to reach");   /* some compiler require default to do something */
     }
 }
 
@@ -1012,6 +1261,10 @@ static size_t ZSTD_refDictContent(ZSTD_DCtx* dctx, const void* dict, size_t dict
     dctx->virtualStart = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->prefixStart));
     dctx->prefixStart = dict;
     dctx->previousDstEnd = (const char*)dict + dictSize;
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    dctx->dictContentBeginForFuzzing = dctx->prefixStart;
+    dctx->dictContentEndForFuzzing = dctx->previousDstEnd;
+#endif
     return 0;
 }
 
@@ -1025,7 +1278,7 @@ ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
     const BYTE* dictPtr = (const BYTE*)dict;
     const BYTE* const dictEnd = dictPtr + dictSize;
 
-    RETURN_ERROR_IF(dictSize <= 8, dictionary_corrupted);
+    RETURN_ERROR_IF(dictSize <= 8, dictionary_corrupted, "dict is too small");
     assert(MEM_readLE32(dict) == ZSTD_MAGIC_DICTIONARY);   /* dict must be valid */
     dictPtr += 8;   /* skip header = magic + dictID */
 
@@ -1041,63 +1294,69 @@ ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
                                                 workspace, workspaceSize);
 #else
         size_t const hSize = HUF_readDTableX2_wksp(entropy->hufTable,
-                                                dictPtr, dictEnd - dictPtr,
+                                                dictPtr, (size_t)(dictEnd - dictPtr),
                                                 workspace, workspaceSize);
 #endif
-        RETURN_ERROR_IF(HUF_isError(hSize), dictionary_corrupted);
+        RETURN_ERROR_IF(HUF_isError(hSize), dictionary_corrupted, "");
         dictPtr += hSize;
     }
 
     {   short offcodeNCount[MaxOff+1];
         unsigned offcodeMaxValue = MaxOff, offcodeLog;
-        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
-        RETURN_ERROR_IF(FSE_isError(offcodeHeaderSize), dictionary_corrupted);
-        RETURN_ERROR_IF(offcodeMaxValue > MaxOff, dictionary_corrupted);
-        RETURN_ERROR_IF(offcodeLog > OffFSELog, dictionary_corrupted);
+        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, (size_t)(dictEnd-dictPtr));
+        RETURN_ERROR_IF(FSE_isError(offcodeHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(offcodeMaxValue > MaxOff, dictionary_corrupted, "");
+        RETURN_ERROR_IF(offcodeLog > OffFSELog, dictionary_corrupted, "");
         ZSTD_buildFSETable( entropy->OFTable,
                             offcodeNCount, offcodeMaxValue,
                             OF_base, OF_bits,
-                            offcodeLog);
+                            offcodeLog,
+                            entropy->workspace, sizeof(entropy->workspace),
+                            /* bmi2 */0);
         dictPtr += offcodeHeaderSize;
     }
 
     {   short matchlengthNCount[MaxML+1];
         unsigned matchlengthMaxValue = MaxML, matchlengthLog;
-        size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
-        RETURN_ERROR_IF(FSE_isError(matchlengthHeaderSize), dictionary_corrupted);
-        RETURN_ERROR_IF(matchlengthMaxValue > MaxML, dictionary_corrupted);
-        RETURN_ERROR_IF(matchlengthLog > MLFSELog, dictionary_corrupted);
+        size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, (size_t)(dictEnd-dictPtr));
+        RETURN_ERROR_IF(FSE_isError(matchlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(matchlengthMaxValue > MaxML, dictionary_corrupted, "");
+        RETURN_ERROR_IF(matchlengthLog > MLFSELog, dictionary_corrupted, "");
         ZSTD_buildFSETable( entropy->MLTable,
                             matchlengthNCount, matchlengthMaxValue,
                             ML_base, ML_bits,
-                            matchlengthLog);
+                            matchlengthLog,
+                            entropy->workspace, sizeof(entropy->workspace),
+                            /* bmi2 */ 0);
         dictPtr += matchlengthHeaderSize;
     }
 
     {   short litlengthNCount[MaxLL+1];
         unsigned litlengthMaxValue = MaxLL, litlengthLog;
-        size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
-        RETURN_ERROR_IF(FSE_isError(litlengthHeaderSize), dictionary_corrupted);
-        RETURN_ERROR_IF(litlengthMaxValue > MaxLL, dictionary_corrupted);
-        RETURN_ERROR_IF(litlengthLog > LLFSELog, dictionary_corrupted);
+        size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, (size_t)(dictEnd-dictPtr));
+        RETURN_ERROR_IF(FSE_isError(litlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(litlengthMaxValue > MaxLL, dictionary_corrupted, "");
+        RETURN_ERROR_IF(litlengthLog > LLFSELog, dictionary_corrupted, "");
         ZSTD_buildFSETable( entropy->LLTable,
                             litlengthNCount, litlengthMaxValue,
                             LL_base, LL_bits,
-                            litlengthLog);
+                            litlengthLog,
+                            entropy->workspace, sizeof(entropy->workspace),
+                            /* bmi2 */ 0);
         dictPtr += litlengthHeaderSize;
     }
 
-    RETURN_ERROR_IF(dictPtr+12 > dictEnd, dictionary_corrupted);
+    RETURN_ERROR_IF(dictPtr+12 > dictEnd, dictionary_corrupted, "");
     {   int i;
         size_t const dictContentSize = (size_t)(dictEnd - (dictPtr+12));
         for (i=0; i<3; i++) {
             U32 const rep = MEM_readLE32(dictPtr); dictPtr += 4;
-            RETURN_ERROR_IF(rep==0 || rep >= dictContentSize,
-                            dictionary_corrupted);
+            RETURN_ERROR_IF(rep==0 || rep > dictContentSize,
+                            dictionary_corrupted, "");
             entropy->rep[i] = rep;
     }   }
 
-    return dictPtr - (const BYTE*)dict;
+    return (size_t)(dictPtr - (const BYTE*)dict);
 }
 
 static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
@@ -1111,7 +1370,7 @@ static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict
 
     /* load entropy tables */
     {   size_t const eSize = ZSTD_loadDEntropy(&dctx->entropy, dict, dictSize);
-        RETURN_ERROR_IF(ZSTD_isError(eSize), dictionary_corrupted);
+        RETURN_ERROR_IF(ZSTD_isError(eSize), dictionary_corrupted, "");
         dict = (const char*)dict + eSize;
         dictSize -= eSize;
     }
@@ -1124,8 +1383,12 @@ static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict
 size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
 {
     assert(dctx != NULL);
+#if ZSTD_TRACE
+    dctx->traceCtx = ZSTD_trace_decompress_begin(dctx);
+#endif
     dctx->expected = ZSTD_startingInputLength(dctx->format);  /* dctx->format must be properly set */
     dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->processedCSize = 0;
     dctx->decodedSize = 0;
     dctx->previousDstEnd = NULL;
     dctx->prefixStart = NULL;
@@ -1134,8 +1397,9 @@ size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
     dctx->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
     dctx->litEntropy = dctx->fseEntropy = 0;
     dctx->dictID = 0;
+    dctx->bType = bt_reserved;
     ZSTD_STATIC_ASSERT(sizeof(dctx->entropy.rep) == sizeof(repStartValue));
-    memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
+    ZSTD_memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
     dctx->LLTptr = dctx->entropy.LLTable;
     dctx->MLTptr = dctx->entropy.MLTable;
     dctx->OFTptr = dctx->entropy.OFTable;
@@ -1145,11 +1409,11 @@ size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
 
 size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
 {
-    FORWARD_IF_ERROR( ZSTD_decompressBegin(dctx) );
+    FORWARD_IF_ERROR( ZSTD_decompressBegin(dctx) , "");
     if (dict && dictSize)
         RETURN_ERROR_IF(
             ZSTD_isError(ZSTD_decompress_insertDictionary(dctx, dict, dictSize)),
-            dictionary_corrupted);
+            dictionary_corrupted, "");
     return 0;
 }
 
@@ -1168,7 +1432,7 @@ size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
         DEBUGLOG(4, "DDict is %s",
                     dctx->ddictIsCold ? "~cold~" : "hot!");
     }
-    FORWARD_IF_ERROR( ZSTD_decompressBegin(dctx) );
+    FORWARD_IF_ERROR( ZSTD_decompressBegin(dctx) , "");
     if (ddict) {   /* NULL ddict is equivalent to no dictionary */
         ZSTD_copyDDictParameters(dctx, ddict);
     }
@@ -1259,11 +1523,11 @@ size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx,
                                          ZSTD_dictLoadMethod_e dictLoadMethod,
                                          ZSTD_dictContentType_e dictContentType)
 {
-    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong);
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
     ZSTD_clearDict(dctx);
-    if (dict && dictSize >= 8) {
+    if (dict && dictSize != 0) {
         dctx->ddictLocal = ZSTD_createDDict_advanced(dict, dictSize, dictLoadMethod, dictContentType, dctx->customMem);
-        RETURN_ERROR_IF(dctx->ddictLocal == NULL, memory_allocation);
+        RETURN_ERROR_IF(dctx->ddictLocal == NULL, memory_allocation, "NULL pointer!");
         dctx->ddict = dctx->ddictLocal;
         dctx->dictUses = ZSTD_use_indefinitely;
     }
@@ -1282,7 +1546,7 @@ size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSi
 
 size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType)
 {
-    FORWARD_IF_ERROR(ZSTD_DCtx_loadDictionary_advanced(dctx, prefix, prefixSize, ZSTD_dlm_byRef, dictContentType));
+    FORWARD_IF_ERROR(ZSTD_DCtx_loadDictionary_advanced(dctx, prefix, prefixSize, ZSTD_dlm_byRef, dictContentType), "");
     dctx->dictUses = ZSTD_use_once;
     return 0;
 }
@@ -1294,14 +1558,14 @@ size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSiz
 
 
 /* ZSTD_initDStream_usingDict() :
- * return : expected size, aka ZSTD_FRAMEHEADERSIZE_PREFIX.
+ * return : expected size, aka ZSTD_startingInputLength().
  * this function cannot fail */
 size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize)
 {
     DEBUGLOG(4, "ZSTD_initDStream_usingDict");
-    FORWARD_IF_ERROR( ZSTD_DCtx_reset(zds, ZSTD_reset_session_only) );
-    FORWARD_IF_ERROR( ZSTD_DCtx_loadDictionary(zds, dict, dictSize) );
-    return ZSTD_FRAMEHEADERSIZE_PREFIX;
+    FORWARD_IF_ERROR( ZSTD_DCtx_reset(zds, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_DCtx_loadDictionary(zds, dict, dictSize) , "");
+    return ZSTD_startingInputLength(zds->format);
 }
 
 /* note : this variant can't fail */
@@ -1316,28 +1580,38 @@ size_t ZSTD_initDStream(ZSTD_DStream* zds)
  * this function cannot fail */
 size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
 {
-    FORWARD_IF_ERROR( ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only) );
-    FORWARD_IF_ERROR( ZSTD_DCtx_refDDict(dctx, ddict) );
-    return ZSTD_FRAMEHEADERSIZE_PREFIX;
+    FORWARD_IF_ERROR( ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_DCtx_refDDict(dctx, ddict) , "");
+    return ZSTD_startingInputLength(dctx->format);
 }
 
 /* ZSTD_resetDStream() :
- * return : expected size, aka ZSTD_FRAMEHEADERSIZE_PREFIX.
+ * return : expected size, aka ZSTD_startingInputLength().
  * this function cannot fail */
 size_t ZSTD_resetDStream(ZSTD_DStream* dctx)
 {
-    FORWARD_IF_ERROR(ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only));
-    return ZSTD_FRAMEHEADERSIZE_PREFIX;
+    FORWARD_IF_ERROR(ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only), "");
+    return ZSTD_startingInputLength(dctx->format);
 }
 
 
 size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
 {
-    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong);
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
     ZSTD_clearDict(dctx);
     if (ddict) {
         dctx->ddict = ddict;
         dctx->dictUses = ZSTD_use_indefinitely;
+        if (dctx->refMultipleDDicts == ZSTD_rmd_refMultipleDDicts) {
+            if (dctx->ddictSet == NULL) {
+                dctx->ddictSet = ZSTD_createDDictHashSet(dctx->customMem);
+                if (!dctx->ddictSet) {
+                    RETURN_ERROR(memory_allocation, "Failed to allocate memory for hash set!");
+                }
+            }
+            assert(!dctx->staticSize);  /* Impossible: ddictSet cannot have been allocated if static dctx */
+            FORWARD_IF_ERROR(ZSTD_DDictHashSet_addDDict(dctx->ddictSet, ddict, dctx->customMem), "");
+        }
     }
     return 0;
 }
@@ -1350,16 +1624,16 @@ size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize)
     ZSTD_bounds const bounds = ZSTD_dParam_getBounds(ZSTD_d_windowLogMax);
     size_t const min = (size_t)1 << bounds.lowerBound;
     size_t const max = (size_t)1 << bounds.upperBound;
-    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong);
-    RETURN_ERROR_IF(maxWindowSize < min, parameter_outOfBound);
-    RETURN_ERROR_IF(maxWindowSize > max, parameter_outOfBound);
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    RETURN_ERROR_IF(maxWindowSize < min, parameter_outOfBound, "");
+    RETURN_ERROR_IF(maxWindowSize > max, parameter_outOfBound, "");
     dctx->maxWindowSize = maxWindowSize;
     return 0;
 }
 
 size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format)
 {
-    return ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, format);
+    return ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, (int)format);
 }
 
 ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
@@ -1375,6 +1649,18 @@ ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
             bounds.upperBound = (int)ZSTD_f_zstd1_magicless;
             ZSTD_STATIC_ASSERT(ZSTD_f_zstd1 < ZSTD_f_zstd1_magicless);
             return bounds;
+        case ZSTD_d_stableOutBuffer:
+            bounds.lowerBound = (int)ZSTD_bm_buffered;
+            bounds.upperBound = (int)ZSTD_bm_stable;
+            return bounds;
+        case ZSTD_d_forceIgnoreChecksum:
+            bounds.lowerBound = (int)ZSTD_d_validateChecksum;
+            bounds.upperBound = (int)ZSTD_d_ignoreChecksum;
+            return bounds;
+        case ZSTD_d_refMultipleDDicts:
+            bounds.lowerBound = (int)ZSTD_rmd_refSingleDDict;
+            bounds.upperBound = (int)ZSTD_rmd_refMultipleDDicts;
+            return bounds;
         default:;
     }
     bounds.error = ERROR(parameter_unsupported);
@@ -1394,12 +1680,35 @@ static int ZSTD_dParam_withinBounds(ZSTD_dParameter dParam, int value)
 }
 
 #define CHECK_DBOUNDS(p,v) {                \
-    RETURN_ERROR_IF(!ZSTD_dParam_withinBounds(p, v), parameter_outOfBound); \
+    RETURN_ERROR_IF(!ZSTD_dParam_withinBounds(p, v), parameter_outOfBound, ""); \
+}
+
+size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value)
+{
+    switch (param) {
+        case ZSTD_d_windowLogMax:
+            *value = (int)ZSTD_highbit32((U32)dctx->maxWindowSize);
+            return 0;
+        case ZSTD_d_format:
+            *value = (int)dctx->format;
+            return 0;
+        case ZSTD_d_stableOutBuffer:
+            *value = (int)dctx->outBufferMode;
+            return 0;
+        case ZSTD_d_forceIgnoreChecksum:
+            *value = (int)dctx->forceIgnoreChecksum;
+            return 0;
+        case ZSTD_d_refMultipleDDicts:
+            *value = (int)dctx->refMultipleDDicts;
+            return 0;
+        default:;
+    }
+    RETURN_ERROR(parameter_unsupported, "");
 }
 
 size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value)
 {
-    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong);
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
     switch(dParam) {
         case ZSTD_d_windowLogMax:
             if (value == 0) value = ZSTD_WINDOWLOG_LIMIT_DEFAULT;
@@ -1410,9 +1719,24 @@ size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value
             CHECK_DBOUNDS(ZSTD_d_format, value);
             dctx->format = (ZSTD_format_e)value;
             return 0;
+        case ZSTD_d_stableOutBuffer:
+            CHECK_DBOUNDS(ZSTD_d_stableOutBuffer, value);
+            dctx->outBufferMode = (ZSTD_bufferMode_e)value;
+            return 0;
+        case ZSTD_d_forceIgnoreChecksum:
+            CHECK_DBOUNDS(ZSTD_d_forceIgnoreChecksum, value);
+            dctx->forceIgnoreChecksum = (ZSTD_forceIgnoreChecksum_e)value;
+            return 0;
+        case ZSTD_d_refMultipleDDicts:
+            CHECK_DBOUNDS(ZSTD_d_refMultipleDDicts, value);
+            if (dctx->staticSize != 0) {
+                RETURN_ERROR(parameter_unsupported, "Static dctx does not support multiple DDicts!");
+            }
+            dctx->refMultipleDDicts = (ZSTD_refMultipleDDicts_e)value;
+            return 0;
         default:;
     }
-    RETURN_ERROR(parameter_unsupported);
+    RETURN_ERROR(parameter_unsupported, "");
 }
 
 size_t ZSTD_DCtx_reset(ZSTD_DCtx* dctx, ZSTD_ResetDirective reset)
@@ -1424,10 +1748,9 @@ size_t ZSTD_DCtx_reset(ZSTD_DCtx* dctx, ZSTD_ResetDirective reset)
     }
     if ( (reset == ZSTD_reset_parameters)
       || (reset == ZSTD_reset_session_and_parameters) ) {
-        RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong);
+        RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
         ZSTD_clearDict(dctx);
-        dctx->format = ZSTD_f_zstd1;
-        dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
+        ZSTD_DCtx_resetParameters(dctx);
     }
     return 0;
 }
@@ -1445,7 +1768,7 @@ size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long
     unsigned long long const neededSize = MIN(frameContentSize, neededRBSize);
     size_t const minRBSize = (size_t) neededSize;
     RETURN_ERROR_IF((unsigned long long)minRBSize != neededSize,
-                    frameParameter_windowTooLarge);
+                    frameParameter_windowTooLarge, "");
     return minRBSize;
 }
 
@@ -1463,30 +1786,94 @@ size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize)
     ZSTD_frameHeader zfh;
     size_t const err = ZSTD_getFrameHeader(&zfh, src, srcSize);
     if (ZSTD_isError(err)) return err;
-    RETURN_ERROR_IF(err>0, srcSize_wrong);
+    RETURN_ERROR_IF(err>0, srcSize_wrong, "");
     RETURN_ERROR_IF(zfh.windowSize > windowSizeMax,
-                    frameParameter_windowTooLarge);
+                    frameParameter_windowTooLarge, "");
     return ZSTD_estimateDStreamSize((size_t)zfh.windowSize);
 }
 
 
 /* *****   Decompression   ***** */
 
-MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+static int ZSTD_DCtx_isOverflow(ZSTD_DStream* zds, size_t const neededInBuffSize, size_t const neededOutBuffSize)
 {
-    size_t const length = MIN(dstCapacity, srcSize);
-    memcpy(dst, src, length);
-    return length;
+    return (zds->inBuffSize + zds->outBuffSize) >= (neededInBuffSize + neededOutBuffSize) * ZSTD_WORKSPACETOOLARGE_FACTOR;
 }
 
+static void ZSTD_DCtx_updateOversizedDuration(ZSTD_DStream* zds, size_t const neededInBuffSize, size_t const neededOutBuffSize)
+{
+    if (ZSTD_DCtx_isOverflow(zds, neededInBuffSize, neededOutBuffSize))
+        zds->oversizedDuration++;
+    else
+        zds->oversizedDuration = 0;
+}
+
+static int ZSTD_DCtx_isOversizedTooLong(ZSTD_DStream* zds)
+{
+    return zds->oversizedDuration >= ZSTD_WORKSPACETOOLARGE_MAXDURATION;
+}
+
+/* Checks that the output buffer hasn't changed if ZSTD_obm_stable is used. */
+static size_t ZSTD_checkOutBuffer(ZSTD_DStream const* zds, ZSTD_outBuffer const* output)
+{
+    ZSTD_outBuffer const expect = zds->expectedOutBuffer;
+    /* No requirement when ZSTD_obm_stable is not enabled. */
+    if (zds->outBufferMode != ZSTD_bm_stable)
+        return 0;
+    /* Any buffer is allowed in zdss_init, this must be the same for every other call until
+     * the context is reset.
+     */
+    if (zds->streamStage == zdss_init)
+        return 0;
+    /* The buffer must match our expectation exactly. */
+    if (expect.dst == output->dst && expect.pos == output->pos && expect.size == output->size)
+        return 0;
+    RETURN_ERROR(dstBuffer_wrong, "ZSTD_d_stableOutBuffer enabled but output differs!");
+}
+
+/* Calls ZSTD_decompressContinue() with the right parameters for ZSTD_decompressStream()
+ * and updates the stage and the output buffer state. This call is extracted so it can be
+ * used both when reading directly from the ZSTD_inBuffer, and in buffered input mode.
+ * NOTE: You must break after calling this function since the streamStage is modified.
+ */
+static size_t ZSTD_decompressContinueStream(
+            ZSTD_DStream* zds, char** op, char* oend,
+            void const* src, size_t srcSize) {
+    int const isSkipFrame = ZSTD_isSkipFrame(zds);
+    if (zds->outBufferMode == ZSTD_bm_buffered) {
+        size_t const dstSize = isSkipFrame ? 0 : zds->outBuffSize - zds->outStart;
+        size_t const decodedSize = ZSTD_decompressContinue(zds,
+                zds->outBuff + zds->outStart, dstSize, src, srcSize);
+        FORWARD_IF_ERROR(decodedSize, "");
+        if (!decodedSize && !isSkipFrame) {
+            zds->streamStage = zdss_read;
+        } else {
+            zds->outEnd = zds->outStart + decodedSize;
+            zds->streamStage = zdss_flush;
+        }
+    } else {
+        /* Write directly into the output buffer */
+        size_t const dstSize = isSkipFrame ? 0 : (size_t)(oend - *op);
+        size_t const decodedSize = ZSTD_decompressContinue(zds, *op, dstSize, src, srcSize);
+        FORWARD_IF_ERROR(decodedSize, "");
+        *op += decodedSize;
+        /* Flushing is not needed. */
+        zds->streamStage = zdss_read;
+        assert(*op <= oend);
+        assert(zds->outBufferMode == ZSTD_bm_stable);
+    }
+    return 0;
+}
 
 size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
 {
-    const char* const istart = (const char*)(input->src) + input->pos;
-    const char* const iend = (const char*)(input->src) + input->size;
+    const char* const src = (const char*)input->src;
+    const char* const istart = input->pos != 0 ? src + input->pos : src;
+    const char* const iend = input->size != 0 ? src + input->size : src;
     const char* ip = istart;
-    char* const ostart = (char*)(output->dst) + output->pos;
-    char* const oend = (char*)(output->dst) + output->size;
+    char* const dst = (char*)output->dst;
+    char* const ostart = output->pos != 0 ? dst + output->pos : dst;
+    char* const oend = output->size != 0 ? dst + output->size : dst;
     char* op = ostart;
     U32 someMoreWork = 1;
 
@@ -1502,6 +1889,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
         "forbidden. out: pos: %u   vs size: %u",
         (U32)output->pos, (U32)output->size);
     DEBUGLOG(5, "input size : %u", (U32)(input->size - input->pos));
+    FORWARD_IF_ERROR(ZSTD_checkOutBuffer(zds, output), "");
 
     while (someMoreWork) {
         switch(zds->streamStage)
@@ -1512,6 +1900,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             zds->lhSize = zds->inPos = zds->outStart = zds->outEnd = 0;
             zds->legacyVersion = 0;
             zds->hostageByte = 0;
+            zds->expectedOutBuffer = *output;
             /* fall-through */
 
         case zdss_loadHeader :
@@ -1526,6 +1915,9 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             }   }
 #endif
             {   size_t const hSize = ZSTD_getFrameHeader_advanced(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format);
+                if (zds->refMultipleDDicts && zds->ddictSet) {
+                    ZSTD_DCtx_selectFrameDDict(zds);
+                }
                 DEBUGLOG(5, "header size : %u", (U32)hSize);
                 if (ZSTD_isError(hSize)) {
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
@@ -1539,7 +1931,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                             "legacy support is incompatible with static dctx");
                         FORWARD_IF_ERROR(ZSTD_initLegacyStream(&zds->legacyContext,
                                     zds->previousLegacyVersion, legacyVersion,
-                                    dict, dictSize));
+                                    dict, dictSize), "");
                         zds->legacyVersion = zds->previousLegacyVersion = legacyVersion;
                         {   size_t const hint = ZSTD_decompressLegacyStream(zds->legacyContext, legacyVersion, output, input);
                             if (hint==0) zds->streamStage = zdss_init;   /* or stay in stage zdss_loadHeader */
@@ -1554,24 +1946,25 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                     assert(iend >= ip);
                     if (toLoad > remainingInput) {   /* not enough input to load full header */
                         if (remainingInput > 0) {
-                            memcpy(zds->headerBuffer + zds->lhSize, ip, remainingInput);
+                            ZSTD_memcpy(zds->headerBuffer + zds->lhSize, ip, remainingInput);
                             zds->lhSize += remainingInput;
                         }
                         input->pos = input->size;
-                        return (MAX(ZSTD_FRAMEHEADERSIZE_MIN, hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
+                        return (MAX((size_t)ZSTD_FRAMEHEADERSIZE_MIN(zds->format), hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
                     }
                     assert(ip != NULL);
-                    memcpy(zds->headerBuffer + zds->lhSize, ip, toLoad); zds->lhSize = hSize; ip += toLoad;
+                    ZSTD_memcpy(zds->headerBuffer + zds->lhSize, ip, toLoad); zds->lhSize = hSize; ip += toLoad;
                     break;
             }   }
 
             /* check for single-pass mode opportunity */
-            if (zds->fParams.frameContentSize && zds->fParams.windowSize /* skippable frame if == 0 */
+            if (zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
+                && zds->fParams.frameType != ZSTD_skippableFrame
                 && (U64)(size_t)(oend-op) >= zds->fParams.frameContentSize) {
-                size_t const cSize = ZSTD_findFrameCompressedSize(istart, iend-istart);
+                size_t const cSize = ZSTD_findFrameCompressedSize(istart, (size_t)(iend-istart));
                 if (cSize <= (size_t)(iend-istart)) {
                     /* shortcut : using single-pass mode */
-                    size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, oend-op, istart, cSize, ZSTD_getDDict(zds));
+                    size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, (size_t)(oend-op), istart, cSize, ZSTD_getDDict(zds));
                     if (ZSTD_isError(decompressedSize)) return decompressedSize;
                     DEBUGLOG(4, "shortcut to single-pass ZSTD_decompress_usingDDict()")
                     ip = istart + cSize;
@@ -1582,15 +1975,23 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                     break;
             }   }
 
+            /* Check output buffer is large enough for ZSTD_odm_stable. */
+            if (zds->outBufferMode == ZSTD_bm_stable
+                && zds->fParams.frameType != ZSTD_skippableFrame
+                && zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
+                && (U64)(size_t)(oend-op) < zds->fParams.frameContentSize) {
+                RETURN_ERROR(dstSize_tooSmall, "ZSTD_obm_stable passed but ZSTD_outBuffer is too small");
+            }
+
             /* Consume header (see ZSTDds_decodeFrameHeader) */
             DEBUGLOG(4, "Consume header");
-            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDDict(zds, ZSTD_getDDict(zds)));
+            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDDict(zds, ZSTD_getDDict(zds)), "");
 
             if ((MEM_readLE32(zds->headerBuffer) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {  /* skippable frame */
                 zds->expected = MEM_readLE32(zds->headerBuffer + ZSTD_FRAMEIDSIZE);
                 zds->stage = ZSTDds_skipFrame;
             } else {
-                FORWARD_IF_ERROR(ZSTD_decodeFrameHeader(zds, zds->headerBuffer, zds->lhSize));
+                FORWARD_IF_ERROR(ZSTD_decodeFrameHeader(zds, zds->headerBuffer, zds->lhSize), "");
                 zds->expected = ZSTD_blockHeaderSize;
                 zds->stage = ZSTDds_decodeBlockHeader;
             }
@@ -1601,40 +2002,48 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                         (U32)(zds->maxWindowSize >> 10) );
             zds->fParams.windowSize = MAX(zds->fParams.windowSize, 1U << ZSTD_WINDOWLOG_ABSOLUTEMIN);
             RETURN_ERROR_IF(zds->fParams.windowSize > zds->maxWindowSize,
-                            frameParameter_windowTooLarge);
+                            frameParameter_windowTooLarge, "");
 
             /* Adapt buffer sizes to frame header instructions */
             {   size_t const neededInBuffSize = MAX(zds->fParams.blockSizeMax, 4 /* frame checksum */);
-                size_t const neededOutBuffSize = ZSTD_decodingBufferSize_min(zds->fParams.windowSize, zds->fParams.frameContentSize);
-                if ((zds->inBuffSize < neededInBuffSize) || (zds->outBuffSize < neededOutBuffSize)) {
-                    size_t const bufferSize = neededInBuffSize + neededOutBuffSize;
-                    DEBUGLOG(4, "inBuff  : from %u to %u",
-                                (U32)zds->inBuffSize, (U32)neededInBuffSize);
-                    DEBUGLOG(4, "outBuff : from %u to %u",
-                                (U32)zds->outBuffSize, (U32)neededOutBuffSize);
-                    if (zds->staticSize) {  /* static DCtx */
-                        DEBUGLOG(4, "staticSize : %u", (U32)zds->staticSize);
-                        assert(zds->staticSize >= sizeof(ZSTD_DCtx));  /* controlled at init */
-                        RETURN_ERROR_IF(
-                            bufferSize > zds->staticSize - sizeof(ZSTD_DCtx),
-                            memory_allocation);
-                    } else {
-                        ZSTD_free(zds->inBuff, zds->customMem);
-                        zds->inBuffSize = 0;
-                        zds->outBuffSize = 0;
-                        zds->inBuff = (char*)ZSTD_malloc(bufferSize, zds->customMem);
-                        RETURN_ERROR_IF(zds->inBuff == NULL, memory_allocation);
-                    }
-                    zds->inBuffSize = neededInBuffSize;
-                    zds->outBuff = zds->inBuff + zds->inBuffSize;
-                    zds->outBuffSize = neededOutBuffSize;
-            }   }
+                size_t const neededOutBuffSize = zds->outBufferMode == ZSTD_bm_buffered
+                        ? ZSTD_decodingBufferSize_min(zds->fParams.windowSize, zds->fParams.frameContentSize)
+                        : 0;
+
+                ZSTD_DCtx_updateOversizedDuration(zds, neededInBuffSize, neededOutBuffSize);
+
+                {   int const tooSmall = (zds->inBuffSize < neededInBuffSize) || (zds->outBuffSize < neededOutBuffSize);
+                    int const tooLarge = ZSTD_DCtx_isOversizedTooLong(zds);
+
+                    if (tooSmall || tooLarge) {
+                        size_t const bufferSize = neededInBuffSize + neededOutBuffSize;
+                        DEBUGLOG(4, "inBuff  : from %u to %u",
+                                    (U32)zds->inBuffSize, (U32)neededInBuffSize);
+                        DEBUGLOG(4, "outBuff : from %u to %u",
+                                    (U32)zds->outBuffSize, (U32)neededOutBuffSize);
+                        if (zds->staticSize) {  /* static DCtx */
+                            DEBUGLOG(4, "staticSize : %u", (U32)zds->staticSize);
+                            assert(zds->staticSize >= sizeof(ZSTD_DCtx));  /* controlled at init */
+                            RETURN_ERROR_IF(
+                                bufferSize > zds->staticSize - sizeof(ZSTD_DCtx),
+                                memory_allocation, "");
+                        } else {
+                            ZSTD_customFree(zds->inBuff, zds->customMem);
+                            zds->inBuffSize = 0;
+                            zds->outBuffSize = 0;
+                            zds->inBuff = (char*)ZSTD_customMalloc(bufferSize, zds->customMem);
+                            RETURN_ERROR_IF(zds->inBuff == NULL, memory_allocation, "");
+                        }
+                        zds->inBuffSize = neededInBuffSize;
+                        zds->outBuff = zds->inBuff + zds->inBuffSize;
+                        zds->outBuffSize = neededOutBuffSize;
+            }   }   }
             zds->streamStage = zdss_read;
             /* fall-through */
 
         case zdss_read:
             DEBUGLOG(5, "stage zdss_read");
-            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds);
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompressWithInputSize(zds, (size_t)(iend - ip));
                 DEBUGLOG(5, "neededInSize = %u", (U32)neededInSize);
                 if (neededInSize==0) {  /* end of frame */
                     zds->streamStage = zdss_init;
@@ -1642,15 +2051,9 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                     break;
                 }
                 if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
-                    int const isSkipFrame = ZSTD_isSkipFrame(zds);
-                    size_t const decodedSize = ZSTD_decompressContinue(zds,
-                        zds->outBuff + zds->outStart, (isSkipFrame ? 0 : zds->outBuffSize - zds->outStart),
-                        ip, neededInSize);
-                    if (ZSTD_isError(decodedSize)) return decodedSize;
+                    FORWARD_IF_ERROR(ZSTD_decompressContinueStream(zds, &op, oend, ip, neededInSize), "");
                     ip += neededInSize;
-                    if (!decodedSize && !isSkipFrame) break;   /* this was just a header */
-                    zds->outEnd = zds->outStart + decodedSize;
-                    zds->streamStage = zdss_flush;
+                    /* Function modifies the stage so we must break */
                     break;
             }   }
             if (ip==iend) { someMoreWork = 0; break; }   /* no more input */
@@ -1662,33 +2065,29 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                 size_t const toLoad = neededInSize - zds->inPos;
                 int const isSkipFrame = ZSTD_isSkipFrame(zds);
                 size_t loadedSize;
+                /* At this point we shouldn't be decompressing a block that we can stream. */
+                assert(neededInSize == ZSTD_nextSrcSizeToDecompressWithInputSize(zds, iend - ip));
                 if (isSkipFrame) {
                     loadedSize = MIN(toLoad, (size_t)(iend-ip));
                 } else {
                     RETURN_ERROR_IF(toLoad > zds->inBuffSize - zds->inPos,
                                     corruption_detected,
                                     "should never happen");
-                    loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, iend-ip);
+                    loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, (size_t)(iend-ip));
                 }
                 ip += loadedSize;
                 zds->inPos += loadedSize;
                 if (loadedSize < toLoad) { someMoreWork = 0; break; }   /* not enough input, wait for more */
 
                 /* decode loaded input */
-                {   size_t const decodedSize = ZSTD_decompressContinue(zds,
-                        zds->outBuff + zds->outStart, zds->outBuffSize - zds->outStart,
-                        zds->inBuff, neededInSize);
-                    if (ZSTD_isError(decodedSize)) return decodedSize;
-                    zds->inPos = 0;   /* input is consumed */
-                    if (!decodedSize && !isSkipFrame) { zds->streamStage = zdss_read; break; }   /* this was just a header */
-                    zds->outEnd = zds->outStart +  decodedSize;
-            }   }
-            zds->streamStage = zdss_flush;
-            /* fall-through */
-
+                zds->inPos = 0;   /* input is consumed */
+                FORWARD_IF_ERROR(ZSTD_decompressContinueStream(zds, &op, oend, zds->inBuff, neededInSize), "");
+                /* Function modifies the stage so we must break */
+                break;
+            }
         case zdss_flush:
             {   size_t const toFlushSize = zds->outEnd - zds->outStart;
-                size_t const flushedSize = ZSTD_limitCopy(op, oend-op, zds->outBuff + zds->outStart, toFlushSize);
+                size_t const flushedSize = ZSTD_limitCopy(op, (size_t)(oend-op), zds->outBuff + zds->outStart, toFlushSize);
                 op += flushedSize;
                 zds->outStart += flushedSize;
                 if (flushedSize == toFlushSize) {  /* flush completed */
@@ -1708,17 +2107,21 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
 
         default:
             assert(0);    /* impossible */
-            RETURN_ERROR(GENERIC);   /* some compiler require default to do something */
+            RETURN_ERROR(GENERIC, "impossible to reach");   /* some compiler require default to do something */
     }   }
 
     /* result */
     input->pos = (size_t)(ip - (const char*)(input->src));
     output->pos = (size_t)(op - (char*)(output->dst));
+
+    /* Update the expected output buffer for ZSTD_obm_stable. */
+    zds->expectedOutBuffer = *output;
+
     if ((ip==istart) && (op==ostart)) {  /* no forward progress */
         zds->noForwardProgress ++;
         if (zds->noForwardProgress >= ZSTD_NO_FORWARD_PROGRESS_MAX) {
-            RETURN_ERROR_IF(op==oend, dstSize_tooSmall);
-            RETURN_ERROR_IF(ip==iend, srcSize_wrong);
+            RETURN_ERROR_IF(op==oend, dstSize_tooSmall, "");
+            RETURN_ERROR_IF(ip==iend, srcSize_wrong, "");
             assert(0);
         }
     } else {