leptris 1.9.162.7-aarch64-linux → 1.9.162.9-aarch64-linux
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- checksums.yaml +4 -4
- data/CHANGELOG.md +37 -0
- data/README.adoc +12 -11
- data/Rakefile +7 -1
- data/TODO.perf/01-ext-accelerated-default-reads.md +25 -0
- data/TODO.perf/02-auto-enable-native-layer.md +19 -0
- data/TODO.perf/03-bulk-xpath-result-materialization.md +13 -0
- data/TODO.perf/04-ext-bound-serialization.md +11 -0
- data/TODO.perf/05-native-surface-completion.md +12 -0
- data/TODO.perf/06-perf-battery-native-legs.md +10 -0
- data/ext/leptris/native/native.c +52 -5
- data/lib/leptris/version.rb +1 -1
- data/lib/leptris/xml/element.rb +32 -4
- data/lib/leptris/xml/ffi.rb +49 -9
- data/lib/leptris/xml/native.so +0 -0
- data/lib/leptris/xml/native_layer.rb +19 -18
- data/lib/leptris.rb +15 -0
- metadata +7 -1
checksums.yaml
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
---
|
|
2
2
|
SHA256:
|
|
3
|
-
metadata.gz:
|
|
4
|
-
data.tar.gz:
|
|
3
|
+
metadata.gz: 930a88268698162d62816a60f82f19f5258cb5fd14f2b89106e234823d0a5363
|
|
4
|
+
data.tar.gz: eae4cad381d8792f156210a068c504569b9005963279742d95d842b6750be5ab
|
|
5
5
|
SHA512:
|
|
6
|
-
metadata.gz:
|
|
7
|
-
data.tar.gz:
|
|
6
|
+
metadata.gz: edab3b443d0e7ea7ecf664e23f149c49136f1f6c33244712eadc9254685d0143dcb5d9bcf574f0d7ca61d900fef9c7665a81fe13216bd565d6b3ab9338491733
|
|
7
|
+
data.tar.gz: '012368af4478e4cd8d3d8ba0b5f3789adf5c12c3adf21d939d1c7d87759e74b4c5cd4a475280a64e027f9556f498a4aa1069480352a920677031fae97f0bcfb4'
|
data/CHANGELOG.md
CHANGED
|
@@ -5,6 +5,43 @@ All notable changes to Leptris will be documented in this file.
|
|
|
5
5
|
The format is based on [Keep a Changelog](https://keepachangelog.com/1.0.0/),
|
|
6
6
|
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
|
|
7
7
|
|
|
8
|
+
## [1.9.162.9] - 2026-09-14
|
|
9
|
+
|
|
10
|
+
### Performance
|
|
11
|
+
|
|
12
|
+
- **TODO.perf/01 — ext-accelerated default reads**: with the
|
|
13
|
+
native layer loaded, the DEFAULT binding classes' hot reads
|
|
14
|
+
(`Element#name`, `#content`, `#prefix`, `#[]`'s engine lookups)
|
|
15
|
+
call the ext directly (one C-API dispatch + UTF-8 string build —
|
|
16
|
+
no FFI marshaling): name 276->84ns, attribute 373->199ns per
|
|
17
|
+
read. Warm loops are memo-served and unaffected; the win lands
|
|
18
|
+
on cold one-shot passes (fresh doc, single read per node):
|
|
19
|
+
13-25% on the #187 workload shape. Scope-owned (iterparse)
|
|
20
|
+
elements stay on the FFI path (pool addresses recycle).
|
|
21
|
+
- **TODO.perf/02 — auto-enable**: `require "leptris"` quietly
|
|
22
|
+
activates the native acceleration when the compiled bundle is
|
|
23
|
+
present (platform gems). Falls back to pure FFI when absent
|
|
24
|
+
(ruby-variant on TruffleRuby/JRuby). `LEPTRIS_NO_NATIVE=1`
|
|
25
|
+
forces the FFI path. The explicit `native_layer` require remains
|
|
26
|
+
the parallel-API entry for NativeNode builders.
|
|
27
|
+
|
|
28
|
+
## [1.9.162.8] - 2026-09-14
|
|
29
|
+
|
|
30
|
+
### Added
|
|
31
|
+
|
|
32
|
+
- **Zero-setup TruffleRuby/JRuby (#160)**: the `ruby`-platform gem
|
|
33
|
+
now vendors `libleptris` + `libutf8proc` for arm64/x86_64-darwin
|
|
34
|
+
and x86_64/aarch64-linux (plus musl fallbacks) under
|
|
35
|
+
`lib/leptris/vendor/<platform>/`; `FFI.libleptris_candidates`
|
|
36
|
+
probes them first at require time. Both engines activate and
|
|
37
|
+
load with no system library and no env vars — the release
|
|
38
|
+
assembly gathers the binaries from every platform build leg.
|
|
39
|
+
MRI behavior is unchanged (platform gems resolve first); a
|
|
40
|
+
forced `--platform ruby` MRI install also works via the vendor
|
|
41
|
+
tree. Fixed a split-library hazard along the way: the native
|
|
42
|
+
read layer now resolves the exact image the FFI layer loads
|
|
43
|
+
(two differently-pathed copies split per-document state).
|
|
44
|
+
|
|
8
45
|
## [1.9.162.7] - 2026-09-14
|
|
9
46
|
|
|
10
47
|
### Added
|
data/README.adoc
CHANGED
|
@@ -60,20 +60,21 @@ export LEPTRIS_LIB_PATH=/usr/local/lib/libleptris.dylib
|
|
|
60
60
|
If Leptris can't find the library at startup, every `parse` call raises
|
|
61
61
|
`LoadError`.
|
|
62
62
|
|
|
63
|
-
=== Alternative engines: TruffleRuby and JRuby (#160)
|
|
63
|
+
=== Alternative engines: TruffleRuby and JRuby (#160) — zero setup
|
|
64
64
|
|
|
65
|
-
The
|
|
66
|
-
|
|
67
|
-
|
|
68
|
-
filter out precompiled platform gems at activation and always resolve
|
|
69
|
-
the `ruby` variant, which does not vendor the native library:
|
|
65
|
+
The `ruby`-platform variant (what TruffleRuby and JRuby resolve)
|
|
66
|
+
vendors precompiled `libleptris` + `libutf8proc` binaries for the
|
|
67
|
+
common engine platforms under `lib/leptris/vendor/`:
|
|
70
68
|
|
|
71
|
-
|
|
72
|
-
|
|
73
|
-
|
|
74
|
-
like on MRI, and FFI loads the system library at require time.
|
|
69
|
+
- `arm64-darwin`, `x86_64-darwin`
|
|
70
|
+
- `x86_64-linux`, `aarch64-linux` (glibc; the musl variants are
|
|
71
|
+
tried as fallback for Alpine hosts)
|
|
75
72
|
|
|
76
|
-
|
|
73
|
+
`gem install leptris` just works on both engines — the FFI layer
|
|
74
|
+
selects the matching vendored binary at require time (the binding
|
|
75
|
+
is FFI-based; no MRI C extension anywhere). `LEPTRIS_LIB_PATH`
|
|
76
|
+
still overrides, and a system library remains the fallback for
|
|
77
|
+
OSes the variant does not carry.
|
|
77
78
|
|
|
78
79
|
== Parsing
|
|
79
80
|
|
data/Rakefile
CHANGED
|
@@ -189,9 +189,15 @@ task "gem:native:any" do
|
|
|
189
189
|
sh "rake platform:any gem"
|
|
190
190
|
end
|
|
191
191
|
|
|
192
|
-
desc "Define the gem task to build the
|
|
192
|
+
desc "Define the gem task to build the ruby-platform gem"
|
|
193
193
|
task "platform:any" do
|
|
194
194
|
spec = Gem::Specification::load("leptris.gemspec").dup
|
|
195
|
+
# Zero-setup TruffleRuby/JRuby (#160): per-OS binaries staged
|
|
196
|
+
# under lib/leptris/vendor/<platform>/ by the release assembly
|
|
197
|
+
# (or by hand: cp from each platform build). Absent on a plain
|
|
198
|
+
# checkout — the variant then carries no binaries and the
|
|
199
|
+
# system-library / LEPTRIS_LIB_PATH path applies as before.
|
|
200
|
+
spec.files += Dir.glob("lib/leptris/vendor/**/*")
|
|
195
201
|
task = Gem::PackageTask.new(spec)
|
|
196
202
|
task.define
|
|
197
203
|
end
|
|
@@ -0,0 +1,25 @@
|
|
|
1
|
+
# 01 — Ext-accelerated default reads (no FFI marshaling on hot paths)
|
|
2
|
+
|
|
3
|
+
Status: SHIPPED (1.9.162.9) — name/content/prefix/[]-cold-path ride
|
|
4
|
+
the ext when the bundle is loaded. Per-read 276->84ns (3.3x,
|
|
5
|
+
pre-address) / attr 373->199ns. HONEST GATE FINDING: warm loops
|
|
6
|
+
cannot see it (memos serve after first touch) — the gate is the
|
|
7
|
+
COLD one-shot pass (fresh doc, one read per node): 13-25% on the
|
|
8
|
+
#187 shape. UTF-8 encoding via rb_utf8_str_new_cstr (suite specs
|
|
9
|
+
run both paths in one process: native_layer_spec sets the flag
|
|
10
|
+
mid-suite). Remaining: children bulk for binding classes.
|
|
11
|
+
|
|
12
|
+
The default binding's hot reads each pay FFI per-call marshaling
|
|
13
|
+
(:string returns allocate through FFI, args marshal through the
|
|
14
|
+
boundary — ~200-500ns vs ~50ns for a direct ext call). The native
|
|
15
|
+
bundle already dlsyms every reader; add address-based C functions
|
|
16
|
+
the DEFAULT binding classes call when the bundle is loaded:
|
|
17
|
+
|
|
18
|
+
- Element#name / #content / #[]
|
|
19
|
+
- Node#children / #element_children (bulk: one C pass allocating
|
|
20
|
+
binding wrappers — the bulk_wrap shape, 1.27-1.30x measured)
|
|
21
|
+
- Element#expanded_name (three out-params today)
|
|
22
|
+
|
|
23
|
+
Gate: suite green with AND without the bundle; drift-record rows
|
|
24
|
+
(attr reads 1.88x, text 6.18x) must improve; cold-walk gate must
|
|
25
|
+
not regress (cold-walk-regression-gate memory).
|
|
@@ -0,0 +1,19 @@
|
|
|
1
|
+
# 02 — Auto-enable the native layer when the bundle is present
|
|
2
|
+
|
|
3
|
+
Status: SHIPPED (1.9.162.9) — require "leptris" quietly
|
|
4
|
+
probes for the native bundle and activates when present.
|
|
5
|
+
LEPTRIS_NO_NATIVE=1 forces the FFI path (suite exercises both).
|
|
6
|
+
The explicit native_layer require remains the parallel-API entry
|
|
7
|
+
for NativeNode builders.
|
|
8
|
+
|
|
9
|
+
Today `require "leptris/xml/native_layer"` is opt-in. End-state:
|
|
10
|
+
`require "leptris"` quietly probes for the vendored native.bundle
|
|
11
|
+
and, when present, activates the acceleration (01 makes the
|
|
12
|
+
default classes benefit). Silent fallback to FFI when absent
|
|
13
|
+
(ruby-variant on TruffleRuby/JRuby, custom builds). The explicit
|
|
14
|
+
native_layer require remains the parallel-API entry
|
|
15
|
+
(NativeNode builders).
|
|
16
|
+
|
|
17
|
+
Requires: 01 (so auto-enable accelerates the DEFAULT API, not a
|
|
18
|
+
parallel one), and a divergence test — same specs pass in both
|
|
19
|
+
modes (the suite already runs both; make it explicit).
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
# 03 — Bulk XPath result materialization
|
|
2
|
+
|
|
3
|
+
Status: TODO
|
|
4
|
+
|
|
5
|
+
NodeSet#fetch materializes results per node (wrap per item +
|
|
6
|
+
ResultText/ResultAttr captures). The #185 profile counted
|
|
7
|
+
NodeSet#initialize at 2.8% and Class#new 11.7% of walk wall —
|
|
8
|
+
result materialization shares that class of cost. Batch via the
|
|
9
|
+
ext: one C pass over leptris_xpath_result_get_node(i) building
|
|
10
|
+
the Ruby array with kinds, capturing text values in C.
|
|
11
|
+
|
|
12
|
+
Gate: xpath-heavy specs unchanged; xpath row (28,878 ips small)
|
|
13
|
+
improves on the battery.
|
|
@@ -0,0 +1,11 @@
|
|
|
1
|
+
# 04 — Serialization buffer cycle in C
|
|
2
|
+
|
|
3
|
+
Status: TODO
|
|
4
|
+
|
|
5
|
+
FFI.serialize_into_string does the two-call sized protocol
|
|
6
|
+
(needed probe, then copy) with Ruby-side buffer management per
|
|
7
|
+
to_xml call. Bind the cycle in the ext: one C function taking the
|
|
8
|
+
ffi_function pointer + node address, returning a Ruby String.
|
|
9
|
+
|
|
10
|
+
Gate: serialize row (3.24x baseline) improves; c14n specs
|
|
11
|
+
byte-identical.
|
|
@@ -0,0 +1,12 @@
|
|
|
1
|
+
# 05 — Complete the NativeNode read surface
|
|
2
|
+
|
|
3
|
+
Status: TODO
|
|
4
|
+
|
|
5
|
+
Missing on NativeNode vs Element: #attributes (bulk attr hash),
|
|
6
|
+
#attribute_nodes, #namespaces / #prefix, #line / #byte_offset,
|
|
7
|
+
#inner_html / #to_xml (delegate to binding serialization), write
|
|
8
|
+
faces beyond add_child (set_attribute, set_text, remove). Reads
|
|
9
|
+
first; writes stay FFI (cold) unless measured otherwise.
|
|
10
|
+
|
|
11
|
+
Gate: native_layer specs mirror the binding feature specs;
|
|
12
|
+
descriptor_walk parity where applicable.
|
|
@@ -0,0 +1,10 @@
|
|
|
1
|
+
# 06 — Native-layer legs in the perf battery + drift record
|
|
2
|
+
|
|
3
|
+
Status: TODO
|
|
4
|
+
|
|
5
|
+
benchmark/ gains native-layer legs (walk/build/attr rows via
|
|
6
|
+
NativeNode beside the binding rows) so the drift record tracks
|
|
7
|
+
both surfaces; TODO.restructure/17's record gains a native table.
|
|
8
|
+
|
|
9
|
+
Gate: one command (rake benchmark:all or the two scripts) emits
|
|
10
|
+
both tables; load-gated per the shared-machine discipline.
|
data/ext/leptris/native/native.c
CHANGED
|
@@ -21,6 +21,7 @@ typedef int (*children_ex_fn)(void *, void **, int *, int);
|
|
|
21
21
|
typedef int (*node_type_fn)(void *);
|
|
22
22
|
typedef void *(*next_sibling_fn)(void *);
|
|
23
23
|
typedef void *(*parent_fn)(void *);
|
|
24
|
+
typedef const char *(*elem_prefix_fn)(void *);
|
|
24
25
|
typedef const char *(*element_text_fn)(void *);
|
|
25
26
|
typedef void *(*doc_create_fn)(void);
|
|
26
27
|
typedef void *(*elem_create_fn)(void *, const char *);
|
|
@@ -37,6 +38,7 @@ static children_ex_fn f_children_ex;
|
|
|
37
38
|
static node_type_fn f_node_type;
|
|
38
39
|
static next_sibling_fn f_next_sibling;
|
|
39
40
|
static parent_fn f_parent;
|
|
41
|
+
static elem_prefix_fn f_elem_prefix;
|
|
40
42
|
static element_text_fn f_element_text;
|
|
41
43
|
static doc_create_fn f_doc_create;
|
|
42
44
|
static elem_create_fn f_elem_create;
|
|
@@ -94,6 +96,7 @@ static void resolve_symbols(const char *lib_path)
|
|
|
94
96
|
f_node_type = (node_type_fn)lib_sym(h, "leptris_node_get_type");
|
|
95
97
|
f_next_sibling = (next_sibling_fn)lib_sym(h, "leptris_node_next_sibling");
|
|
96
98
|
f_parent = (parent_fn)lib_sym(h, "leptris_node_parent");
|
|
99
|
+
f_elem_prefix = (elem_prefix_fn)lib_sym(h, "leptris_element_prefix");
|
|
97
100
|
f_element_text = (element_text_fn)lib_sym(h, "leptris_element_text");
|
|
98
101
|
f_doc_create = (doc_create_fn)lib_sym(h, "leptris_document_create");
|
|
99
102
|
f_elem_create = (elem_create_fn)lib_sym(h, "leptris_element_create");
|
|
@@ -105,7 +108,8 @@ static void resolve_symbols(const char *lib_path)
|
|
|
105
108
|
if (!f_elem_name || !f_text_content || !f_attr ||
|
|
106
109
|
!f_children_ex || !f_node_type || !f_next_sibling || !f_parent ||
|
|
107
110
|
!f_element_text || !f_doc_create || !f_elem_create || !f_text_create ||
|
|
108
|
-
!f_create_child || !f_append_child || !f_set_root || !f_doc_free
|
|
111
|
+
!f_create_child || !f_append_child || !f_set_root || !f_doc_free ||
|
|
112
|
+
!f_elem_prefix)
|
|
109
113
|
rb_raise(rb_eRuntimeError, "libleptris symbols missing");
|
|
110
114
|
}
|
|
111
115
|
|
|
@@ -115,7 +119,7 @@ static VALUE nn_name(VALUE self)
|
|
|
115
119
|
const char *s;
|
|
116
120
|
TypedData_Get_Struct(self, struct native_node, &nn_type, n);
|
|
117
121
|
s = f_elem_name(n->ptr);
|
|
118
|
-
return s ?
|
|
122
|
+
return s ? rb_utf8_str_new_cstr(s) : Qnil;
|
|
119
123
|
}
|
|
120
124
|
|
|
121
125
|
static VALUE nn_content(VALUE self)
|
|
@@ -127,7 +131,7 @@ static VALUE nn_content(VALUE self)
|
|
|
127
131
|
* nodes carry it directly. */
|
|
128
132
|
s = f_node_type(n->ptr) == 0 ? f_element_text(n->ptr)
|
|
129
133
|
: f_text_content(n->ptr);
|
|
130
|
-
return s ?
|
|
134
|
+
return s ? rb_utf8_str_new_cstr(s) : Qnil;
|
|
131
135
|
}
|
|
132
136
|
|
|
133
137
|
static VALUE nn_attribute(VALUE self, VALUE name)
|
|
@@ -136,7 +140,7 @@ static VALUE nn_attribute(VALUE self, VALUE name)
|
|
|
136
140
|
const char *s;
|
|
137
141
|
TypedData_Get_Struct(self, struct native_node, &nn_type, n);
|
|
138
142
|
s = f_attr(n->ptr, StringValueCStr(name));
|
|
139
|
-
return s ?
|
|
143
|
+
return s ? rb_utf8_str_new_cstr(s) : Qnil;
|
|
140
144
|
}
|
|
141
145
|
|
|
142
146
|
static VALUE nn_allocate(VALUE klass)
|
|
@@ -355,6 +359,43 @@ static VALUE nn_address(VALUE self)
|
|
|
355
359
|
return ULL2NUM((uint64_t)(uintptr_t)n->ptr);
|
|
356
360
|
}
|
|
357
361
|
|
|
362
|
+
/* ---- Address-based fast readers (TODO.perf/01) -----------------
|
|
363
|
+
* The DEFAULT binding classes call these when the bundle is
|
|
364
|
+
* loaded: one C-API dispatch + rb_utf8_str_new_cstr — no FFI
|
|
365
|
+
* marshaling. Addresses come from FFI::Pointer#address (a Ruby
|
|
366
|
+
* ivar read on the ffi gem's Pointer). */
|
|
367
|
+
static VALUE nf_name(VALUE self, VALUE addr)
|
|
368
|
+
{
|
|
369
|
+
const char *s;
|
|
370
|
+
(void)self;
|
|
371
|
+
s = f_elem_name((void *)(uintptr_t)NUM2ULL(addr));
|
|
372
|
+
return s ? rb_utf8_str_new_cstr(s) : Qnil;
|
|
373
|
+
}
|
|
374
|
+
|
|
375
|
+
static VALUE nf_element_text(VALUE self, VALUE addr)
|
|
376
|
+
{
|
|
377
|
+
const char *s;
|
|
378
|
+
(void)self;
|
|
379
|
+
s = f_element_text((void *)(uintptr_t)NUM2ULL(addr));
|
|
380
|
+
return s ? rb_utf8_str_new_cstr(s) : Qnil;
|
|
381
|
+
}
|
|
382
|
+
|
|
383
|
+
static VALUE nf_attribute(VALUE self, VALUE addr, VALUE name)
|
|
384
|
+
{
|
|
385
|
+
const char *s;
|
|
386
|
+
(void)self;
|
|
387
|
+
s = f_attr((void *)(uintptr_t)NUM2ULL(addr), StringValueCStr(name));
|
|
388
|
+
return s ? rb_utf8_str_new_cstr(s) : Qnil;
|
|
389
|
+
}
|
|
390
|
+
|
|
391
|
+
static VALUE nf_prefix(VALUE self, VALUE addr)
|
|
392
|
+
{
|
|
393
|
+
const char *s;
|
|
394
|
+
(void)self;
|
|
395
|
+
s = f_elem_prefix((void *)(uintptr_t)NUM2ULL(addr));
|
|
396
|
+
return (s && *s) ? rb_utf8_str_new_cstr(s) : Qnil;
|
|
397
|
+
}
|
|
398
|
+
|
|
358
399
|
static VALUE nn_from(VALUE klass, VALUE document, VALUE element)
|
|
359
400
|
{
|
|
360
401
|
/* Element address comes through the binding's #c_ptr address;
|
|
@@ -383,7 +424,7 @@ static VALUE leptris_native_resolve_rb(VALUE self, VALUE lib_path)
|
|
|
383
424
|
|
|
384
425
|
void Init_native(void)
|
|
385
426
|
{
|
|
386
|
-
VALUE m_leptris, m_xml;
|
|
427
|
+
VALUE m_leptris, m_xml, m_native;
|
|
387
428
|
|
|
388
429
|
m_leptris = rb_define_module("Leptris");
|
|
389
430
|
m_xml = rb_define_module_under(m_leptris, "XML");
|
|
@@ -411,4 +452,10 @@ void Init_native(void)
|
|
|
411
452
|
rb_define_method(c_native_node, "node_type", nn_type_sym, 0);
|
|
412
453
|
rb_define_singleton_method(c_native_node, "resolve!",
|
|
413
454
|
leptris_native_resolve_rb, 1);
|
|
455
|
+
|
|
456
|
+
m_native = rb_define_module_under(m_xml, "Native");
|
|
457
|
+
rb_define_module_function(m_native, "fast_name", nf_name, 1);
|
|
458
|
+
rb_define_module_function(m_native, "fast_element_text", nf_element_text, 1);
|
|
459
|
+
rb_define_module_function(m_native, "fast_attribute", nf_attribute, 2);
|
|
460
|
+
rb_define_module_function(m_native, "fast_prefix", nf_prefix, 1);
|
|
414
461
|
}
|
data/lib/leptris/version.rb
CHANGED
data/lib/leptris/xml/element.rb
CHANGED
|
@@ -1,10 +1,25 @@
|
|
|
1
1
|
# frozen_string_literal: true
|
|
2
2
|
|
|
3
3
|
class Leptris::XML::Element < Leptris::XML::Node
|
|
4
|
+
# TODO.perf/01: with the native bundle loaded, hot reads go
|
|
5
|
+
# through the ext (one C-API dispatch + rb_str_new_cstr — no FFI
|
|
6
|
+
# marshaling; ~3x per read). defined? is near-free when absent.
|
|
7
|
+
# Scope-owned (iterparse-yield) elements stay on the FFI path:
|
|
8
|
+
# their pool addresses recycle across yields and their lifetime
|
|
9
|
+
# rules ride the IterationScope seam — the ext fast path is for
|
|
10
|
+
# document-owned trees.
|
|
11
|
+
def native_fast?
|
|
12
|
+
defined?(Leptris::XML::NATIVE_FAST) && !scope_owned?
|
|
13
|
+
end
|
|
14
|
+
|
|
4
15
|
def name
|
|
5
16
|
return @name if @name
|
|
6
17
|
ensure_alive!
|
|
7
|
-
@name =
|
|
18
|
+
@name = if native_fast?
|
|
19
|
+
Leptris::XML::Native.fast_name(@c_ptr.address)
|
|
20
|
+
else
|
|
21
|
+
Leptris::XML::FFI.leptris_element_name(@c_ptr)
|
|
22
|
+
end
|
|
8
23
|
end
|
|
9
24
|
alias_method :node_name, :name
|
|
10
25
|
|
|
@@ -19,7 +34,11 @@ class Leptris::XML::Element < Leptris::XML::Node
|
|
|
19
34
|
def content
|
|
20
35
|
return @content if memo_hit?(@content_version)
|
|
21
36
|
ensure_alive!
|
|
22
|
-
result =
|
|
37
|
+
result = if native_fast?
|
|
38
|
+
Leptris::XML::Native.fast_element_text(@c_ptr.address)
|
|
39
|
+
else
|
|
40
|
+
Leptris::XML::FFI.leptris_element_text(@c_ptr)
|
|
41
|
+
end
|
|
23
42
|
if @document
|
|
24
43
|
@content = result
|
|
25
44
|
@content_version = @document.version
|
|
@@ -60,7 +79,11 @@ class Leptris::XML::Element < Leptris::XML::Node
|
|
|
60
79
|
# engine call resolves it, then the name is cached.
|
|
61
80
|
return v if !v.nil? || @attributes
|
|
62
81
|
ensure_alive!
|
|
63
|
-
v =
|
|
82
|
+
v = if native_fast?
|
|
83
|
+
Leptris::XML::Native.fast_attribute(@c_ptr.address, name)
|
|
84
|
+
else
|
|
85
|
+
Leptris::XML::FFI.leptris_element_attribute(@c_ptr, name)
|
|
86
|
+
end
|
|
64
87
|
values[name] = v
|
|
65
88
|
return v
|
|
66
89
|
end
|
|
@@ -203,7 +226,12 @@ class Leptris::XML::Element < Leptris::XML::Node
|
|
|
203
226
|
# or nil when the element has none.
|
|
204
227
|
def prefix
|
|
205
228
|
ensure_alive!
|
|
206
|
-
|
|
229
|
+
if native_fast?
|
|
230
|
+
Leptris::XML::Native.fast_prefix(@c_ptr.address)
|
|
231
|
+
else
|
|
232
|
+
result = Leptris::XML::FFI.leptris_element_prefix(@c_ptr)
|
|
233
|
+
result if result && !result.empty?
|
|
234
|
+
end
|
|
207
235
|
end
|
|
208
236
|
|
|
209
237
|
# The written qualified name — "foo:child" with a prefix, the
|
data/lib/leptris/xml/ffi.rb
CHANGED
|
@@ -7,6 +7,50 @@ module Leptris
|
|
|
7
7
|
module FFI
|
|
8
8
|
extend ::FFI::Library
|
|
9
9
|
|
|
10
|
+
# The per-OS vendor directory of the ruby-platform gem
|
|
11
|
+
# (zero-setup TruffleRuby/JRuby, #160): binaries for the
|
|
12
|
+
# common engines' OSes under lib/leptris/vendor/<platform>/.
|
|
13
|
+
# Returns [] in platform gems (no vendor tree) and on
|
|
14
|
+
# OSes the variant does not carry.
|
|
15
|
+
# Dir-name candidates for a RUBY_PLATFORM string, in
|
|
16
|
+
# preference order (glibc before musl — dlopen falls through
|
|
17
|
+
# on musl hosts).
|
|
18
|
+
def self.vendor_platforms_for(ruby_platform)
|
|
19
|
+
case ruby_platform
|
|
20
|
+
when /arm64.*darwin|aarch64.*darwin/ then %w[arm64-darwin]
|
|
21
|
+
when /x86_64.*darwin/ then %w[x86_64-darwin]
|
|
22
|
+
when /aarch64.*linux/ then %w[aarch64-linux aarch64-linux-musl]
|
|
23
|
+
when /x86_64.*linux/ then %w[x86_64-linux x86_64-linux-musl]
|
|
24
|
+
else []
|
|
25
|
+
end
|
|
26
|
+
end
|
|
27
|
+
|
|
28
|
+
def self.vendor_dirs
|
|
29
|
+
vendor_platforms_for(RUBY_PLATFORM)
|
|
30
|
+
.map { |d| File.expand_path("../../leptris/vendor/#{d}", __dir__) }
|
|
31
|
+
.select { |dir| File.directory?(dir) }
|
|
32
|
+
end
|
|
33
|
+
|
|
34
|
+
# THE libleptris candidate list, in order — shared with the
|
|
35
|
+
# native read layer's resolve!. Single source of truth is
|
|
36
|
+
# load-bearing: two differently-pathed copies of the library
|
|
37
|
+
# load as two images with split per-document state.
|
|
38
|
+
def self.libleptris_candidates
|
|
39
|
+
[
|
|
40
|
+
ENV["LEPTRIS_LIB_PATH"],
|
|
41
|
+
*vendor_dirs.flat_map { |d|
|
|
42
|
+
%w[libleptris.dylib libleptris.so libleptris.dll]
|
|
43
|
+
.map { |n| File.join(d, n) }
|
|
44
|
+
},
|
|
45
|
+
File.expand_path("../../libleptris.dylib", __dir__),
|
|
46
|
+
File.expand_path("../../libleptris.so", __dir__),
|
|
47
|
+
File.expand_path("../../libleptris.dll", __dir__),
|
|
48
|
+
"/usr/local/lib/libleptris.dylib",
|
|
49
|
+
"/usr/local/lib/libleptris.so",
|
|
50
|
+
"leptris",
|
|
51
|
+
].compact
|
|
52
|
+
end
|
|
53
|
+
|
|
10
54
|
# libleptris links utf8proc via @rpath (TODO.restructure/20
|
|
11
55
|
# — fn:normalize-unicode). dlopen the VENDORED utf8proc
|
|
12
56
|
# FIRST: once its install name is loaded, dyld resolves
|
|
@@ -17,6 +61,10 @@ module Leptris
|
|
|
17
61
|
# never consults it.
|
|
18
62
|
begin
|
|
19
63
|
ffi_lib [
|
|
64
|
+
*Leptris::XML::FFI.vendor_dirs.flat_map { |d|
|
|
65
|
+
%w[libutf8proc.3.dylib libutf8proc.so.3 libutf8proc.so
|
|
66
|
+
utf8proc.dll].map { |n| File.join(d, n) }
|
|
67
|
+
},
|
|
20
68
|
File.expand_path("../../libutf8proc.3.dylib", __dir__),
|
|
21
69
|
File.expand_path("../../libutf8proc.so.3", __dir__),
|
|
22
70
|
File.expand_path("../../libutf8proc.so", __dir__),
|
|
@@ -29,15 +77,7 @@ module Leptris
|
|
|
29
77
|
# Issue leptris-ruby#49: name the remedy when the library is
|
|
30
78
|
# missing (ruby-platform gem without a vendored libleptris).
|
|
31
79
|
begin
|
|
32
|
-
ffi_lib
|
|
33
|
-
ENV["LEPTRIS_LIB_PATH"],
|
|
34
|
-
File.expand_path("../../libleptris.dylib", __dir__),
|
|
35
|
-
File.expand_path("../../libleptris.so", __dir__),
|
|
36
|
-
File.expand_path("../../libleptris.dll", __dir__),
|
|
37
|
-
"/usr/local/lib/libleptris.dylib",
|
|
38
|
-
"/usr/local/lib/libleptris.so",
|
|
39
|
-
"leptris",
|
|
40
|
-
].compact
|
|
80
|
+
ffi_lib Leptris::XML::FFI.libleptris_candidates
|
|
41
81
|
rescue LoadError => e
|
|
42
82
|
raise LoadError, <<~MSG
|
|
43
83
|
leptris: cannot load the vendored libleptris library.
|
data/lib/leptris/xml/native.so
CHANGED
|
Binary file
|
|
@@ -29,29 +29,30 @@ rescue LoadError => e
|
|
|
29
29
|
end
|
|
30
30
|
|
|
31
31
|
module Leptris::XML::Native
|
|
32
|
-
|
|
33
|
-
|
|
34
|
-
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
].compact.freeze
|
|
32
|
+
# FFI.libleptris_candidates is the single source of truth — the
|
|
33
|
+
# native layer MUST dlsym the exact image the FFI layer loaded,
|
|
34
|
+
# or two library instances split per-document state.
|
|
35
|
+
def self.lib_candidates
|
|
36
|
+
Leptris::XML::FFI.libleptris_candidates
|
|
37
|
+
end
|
|
39
38
|
end
|
|
40
39
|
|
|
41
|
-
|
|
40
|
+
# TODO.perf/01: accelerate the DEFAULT binding classes while the
|
|
41
|
+
# bundle is loaded (name/content/prefix/[] hot reads).
|
|
42
|
+
Leptris::XML::NATIVE_FAST = true
|
|
43
|
+
|
|
44
|
+
candidates = Leptris::XML::Native.lib_candidates
|
|
42
45
|
resolved = candidates.any? do |path|
|
|
43
|
-
|
|
44
|
-
|
|
45
|
-
|
|
46
|
-
|
|
47
|
-
|
|
48
|
-
false
|
|
49
|
-
end
|
|
46
|
+
begin
|
|
47
|
+
Leptris::XML::NativeNode.resolve!(path)
|
|
48
|
+
true
|
|
49
|
+
rescue RuntimeError
|
|
50
|
+
false
|
|
50
51
|
end
|
|
51
|
-
raise LoadError,
|
|
52
|
-
"leptris: native layer loaded but libleptris could not be " \
|
|
53
|
-
"resolved from #{candidates.inspect}" unless resolved
|
|
54
52
|
end
|
|
53
|
+
raise LoadError,
|
|
54
|
+
"leptris: native layer loaded but libleptris could not be " \
|
|
55
|
+
"resolved from #{candidates.inspect}" unless resolved
|
|
55
56
|
|
|
56
57
|
class Leptris::XML::Document
|
|
57
58
|
# Separate identity cache for NativeNodes so binding Node.wrap
|
data/lib/leptris.rb
CHANGED
|
@@ -31,3 +31,18 @@ rescue LoadError => e
|
|
|
31
31
|
(Underlying error: #{e.message})
|
|
32
32
|
MSG
|
|
33
33
|
end
|
|
34
|
+
|
|
35
|
+
# TODO.perf/02: auto-enable the native acceleration layer when the
|
|
36
|
+
# compiled bundle is present (platform gems). Quietly falls back
|
|
37
|
+
# to pure FFI when absent (ruby-variant on TruffleRuby/JRuby,
|
|
38
|
+
# custom builds). LEPTRIS_NO_NATIVE=1 forces the FFI path (the
|
|
39
|
+
# suite uses it to exercise both surfaces). The explicit
|
|
40
|
+
# `require "leptris/xml/native_layer"` remains the parallel-API
|
|
41
|
+
# entry for NativeNode builders.
|
|
42
|
+
unless ENV["LEPTRIS_NO_NATIVE"] == "1"
|
|
43
|
+
begin
|
|
44
|
+
require "leptris/xml/native_layer"
|
|
45
|
+
rescue LoadError
|
|
46
|
+
nil
|
|
47
|
+
end
|
|
48
|
+
end
|
metadata
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
|
2
2
|
name: leptris
|
|
3
3
|
version: !ruby/object:Gem::Version
|
|
4
|
-
version: 1.9.162.
|
|
4
|
+
version: 1.9.162.9
|
|
5
5
|
platform: aarch64-linux
|
|
6
6
|
authors:
|
|
7
7
|
- Ribose Inc.
|
|
@@ -69,6 +69,12 @@ files:
|
|
|
69
69
|
- LICENSE.md
|
|
70
70
|
- README.adoc
|
|
71
71
|
- Rakefile
|
|
72
|
+
- TODO.perf/01-ext-accelerated-default-reads.md
|
|
73
|
+
- TODO.perf/02-auto-enable-native-layer.md
|
|
74
|
+
- TODO.perf/03-bulk-xpath-result-materialization.md
|
|
75
|
+
- TODO.perf/04-ext-bound-serialization.md
|
|
76
|
+
- TODO.perf/05-native-surface-completion.md
|
|
77
|
+
- TODO.perf/06-perf-battery-native-legs.md
|
|
72
78
|
- TODO.restructure/01-constraint-compliance-audit.md
|
|
73
79
|
- TODO.restructure/02-deep-copy-seam.md
|
|
74
80
|
- TODO.restructure/03-evaluation-context-seam.md
|