PyPI - sequence-base - Versions diffs - 0.4.3__tar.gz → 0.4.4__tar.gz - Mend

@@ -0,0 +1,64 @@
+"""
+Correlation id — one value threads a call through sdk → gateway → worker, so "what happened to that
+call" is one grep instead of three log stores and a stopwatch. {from general-sequences obs.py}
+"""
+from __future__ import annotations
+import re
+import uuid
+HEADER = "X-Seq-Request-Id"
+_UNSAFE = re.compile(r"\s+")
+_BREAK = re.compile(r"\r\n|\r|\n")
+_SPACE = re.compile(r"\s")
+_ESCAPED = re.compile(r"\\(.)|_")
+FIELD_MAX = 200
+TEXT_MAX = 4000
+def new_rid() -> str:
+    """Short (read off a terminal, quoted in bug reports) + random (gateway instances don't
+    coordinate, so it must not collide)."""
+    return "r-" + uuid.uuid4().hex[:8]
+def clean(v: object, limit: int = FIELD_MAX) -> str:
+    """Collapse whitespace and cap length. A value with a space would split into two log fields and
+    let anything user-supplied forge keys into the line; 200 chars keeps a real error readable."""
+    s = _UNSAFE.sub("_", str(v))
+    return s if len(s) <= limit else s[:limit - 3] + "..."
+def clean_text(v: object, limit: int = TEXT_MAX) -> str:
+    """A long text - an error with its traceback - as one log field that reads back as it was (unescape_text): a
+    backslash written `\\\\`, an underscore `\\_`, a line break `\\n`, every other whitespace character `_` (read
+    back as a space, so a traceback keeps its indentation), cut at `limit`. It holds no whitespace, so like clean() it
+    can neither split the line nor forge a field into it.
+    WHY escaped and not collapsed: clean() writes a traceback's line breaks and spaces both as `_`, and a reader that
+    turns `_` back into spaces breaks every name with one in it (infer_chunk); a traceback is read for those names."""
+    s = "\\n".join(_SPACE.sub("_", line.replace("\\", "\\\\").replace("_", "\\_")) for line in _BREAK.split(str(v)))
+    if len(s) <= limit:
+        return s
+    s = s[:limit - 3]
+    if (len(s) - len(s.rstrip("\\"))) % 2:
+        s = s[:-1]
+    return s + "..."
+def unescape_text(s: str) -> str:
+    """clean_text's field as the text it was: `\\n` a line break, `\\\\` a backslash, `\\_` an underscore, `_` a
+    space; any other backslash as written. One pass, left to right."""
+    def one(m: re.Match) -> str:
+        if m.group(1) is None:
+            return " "
+        return {"n": "\n", "\\": "\\", "_": "_"}.get(m.group(1), m.group(0))
+    return _ESCAPED.sub(one, s)

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: sequence-base
-Version: 0.4.3
+Version: 0.4.4
 Summary: Shared foundation for the Sequences platform — wire contract, correlation, error taxonomy, constants. Used by the robot SDK, gateway, and container harness.
 Project-URL: Changelog, https://generalsequences.com/docs/changelog/
 Requires-Python: >=3.10

@@ -1,6 +1,6 @@
 [project]
 name = "sequence-base"
-version = "0.4.3"
+version = "0.4.4"
 description = "Shared foundation for the Sequences platform — wire contract, correlation, error taxonomy, constants. Used by the robot SDK, gateway, and container harness."
 requires-python = ">=3.10"
 dependencies = ["pydantic>=2.0"]

@@ -105,7 +105,7 @@ from sequence_base.constants.version import CONTRACT_VERSION
 from sequence_base.obs.log import Timer, enabled, log
-from sequence_base.obs.rid import HEADER, clean, new_rid
+from sequence_base.obs.rid import HEADER, clean, clean_text, new_rid, unescape_text
 from sequence_base.errors.transport import (
@@ -152,7 +152,7 @@ __all__ = [
     "ErrorCode", "ErrorBody", "WarmingError", "ERRORS", "ErrorSpec", "code_for_status", "CONTRACT_VERSION",
-    "HEADER", "new_rid", "clean", "log", "Timer", "enabled",
+    "HEADER", "new_rid", "clean", "clean_text", "unescape_text", "log", "Timer", "enabled",
     "TransportUnavailable", "TransportWarming", "TransportRejected", "TransportOverloaded",

@@ -279,8 +279,8 @@ MAX_WEIGHTS_PARTS = 8
 def check_weights_uri(v: str, what: str = "weights") -> str:
     """A weights origin the build can provision: an official template://, an hf:// repo or an uploaded vol://
-    volume. gs://, s3:// and https:// were removed in contract 0.5.0, after a release that warned for 7 days: files
-    from a bucket or a URL are uploaded once as a volume."""
+    volume. Weights from a storage bucket or a URL were removed in contract 0.5.0, after a release that warned for 7
+    days: their files are uploaded once as a volume."""
     if v.startswith("template://"):
         from sequence_base.contract.templates import parse_template

@@ -9,7 +9,11 @@ import os
 import sys
 import time
-from sequence_base.obs.rid import clean
+from sequence_base.obs.rid import clean, clean_text
+TEXT_FIELDS = frozenset({"error"})
 def enabled() -> bool:
@@ -24,7 +28,7 @@ def log(ev: str, rid: str | None = None, **fields: object) -> None:
     parts = [f"seq ev={clean(ev)}"]
     if rid:
         parts.append(f"rid={clean(rid)}")
-    parts.extend(f"{k}={clean(v)}" for k, v in fields.items() if v is not None)
+    parts.extend(f"{k}={clean_text(v) if k in TEXT_FIELDS else clean(v)}" for k, v in fields.items() if v is not None)
     print(" ".join(parts), file=sys.stdout, flush=True)

sequence-base 0.4.3__tar.gz → 0.4.4__tar.gz

sequence-base 0.4.3tar.gz → 0.4.4tar.gz