Skip to content

[nanvix] E: Drop libm.a from math/cmath/_statistics .so link#7

Closed
esaurez wants to merge 2 commits into
feat/phase1b-tier1-mathmem-sharedfrom
feat/phase1b-drop-libm-from-math-so
Closed

[nanvix] E: Drop libm.a from math/cmath/_statistics .so link#7
esaurez wants to merge 2 commits into
feat/phase1b-tier1-mathmem-sharedfrom
feat/phase1b-drop-libm-from-math-so

Conversation

@esaurez

@esaurez esaurez commented Jun 3, 2026

Copy link
Copy Markdown
Owner

Summary

Phase 1B follow-up. Now that esaurez/nanvix#26 fixes the libm visibility merge in libposix.a (localizes the 28 libm wrapper shadows emitted by Rust's compiler_builtins), python.elf's .dynsym correctly exposes sqrt, cbrt, fma, ceil, fabs, fmod, etc. as GLOBAL DEFAULT. dlopen'd .so modules can now resolve those symbols at runtime against the main executable — exactly like Linux dlopen'd modules resolve them against libm.so.6.

This PR passes --with-libm= (empty) to CPython's configure so the math / cmath / _statistics module link commands no longer include libm.a. Each .so ends up with sqrt / cbrt / etc. as undefined references that the dynamic loader resolves at dlopen time.

Size impact

Module Before After Delta
math.cpython-312.so 532 KB 381 KB -28%
cmath.cpython-312.so 180 KB 91 KB -50%
_statistics.cpython-312.so 28 KB 18 KB -36%

python.elf size is unchanged — libm.a was already linked via --whole-archive into the main binary; that hasn't changed. Net ramfs reduction: ~240 KB across the three modules. The remaining bulk in math.so is CPython wrapper code, not libm.

Architectural pattern

This is the canonical pattern for any Nanvix-on-static-libm CPython extension that uses libm:

  1. Build the extension with no libm in its link command.
  2. The extension carries sqrt / cbrt / etc. as UND symbols.
  3. At dlopen time, the dynamic loader resolves them against python.elf's .dynsym, which has them as GLOBAL DEFAULT because libm.a is in python.elf's --whole-archive LIBS plus the libposix visibility-merge contamination has been fixed at the Nanvix level ([nanvix] Makefile.nanvix: enable ac_cv_func_inet_pton=yes #26).

This matches how Linux dlopen'd CPython math modules resolve libm — through the main executable's .dynsym (Linux uses libm.so.6 instead of the static link, but the resolution mechanism is the same). Future Phase 2/3 modules with libm dependencies (_datetime, _decimal, etc.) automatically benefit from this without needing the libm.a baked into each .so.

Validation

Tested locally on a toolchain image (local-nanvix/toolchain-python:phase0-llfix) bundling esaurez/nanvix#26's patched libposix.a / libnvx_crt0.a:

  • readelf --dyn-syms python.elf | grep sqrt shows GLOBAL DEFAULT sqrt at 0x40415bd0.
  • math.sqrt(4.0) == 2.0 and cmath.sqrt(-1) == 1j round-trip via dlopen.
  • Phase 1B import probe (math, cmath, _statistics, mmap, _contextvars): all 5 modules load via dlopen.
  • Full CPython regrtest: 160/160 modules PASS.
  • Hello + lxml + HTTP smoke tests all pass.

Prerequisites

  • esaurez/nanvix#26 must be merged.
  • The ghcr.io/nanvix/toolchain-python image must be rebuilt to carry the patched libposix.a.

Without these, math.so dlopen would fail with ImportError: symbol not found on sqrt — exactly the failure mode that motivated #26.

Note on the 11 newlib-internal __math_* helpers

__math_invalid, __math_oflow, __math_uflow, __math_divzero, etc. are GLOBAL HIDDEN at source in newlib/libm/common/math_config.h (ported from ARM optimized-routines, same as glibc and musl). They are deliberately library-private and do not appear in python.elf's .dynsym even with this PR — and that is correct, not a bug. They are called only from inside libm's own internals (e.g. __ieee754_sqrt → __math_invalid), which live inside python.elf alongside the helpers themselves; the PC-relative call between them is resolved at static-link time and never touches .dynsym. dlopen'd modules don't need them — they call public names like sqrt, which dispatches to python.elf's sqrt, which internally calls __math_invalid if needed. This is identical to how Linux ships libm.so.6 with the same HIDDEN attribute on the same helpers; dlopen'd Python extensions on Linux never need to look them up either.

Risk

One-line change. Trivially revertable if needed (set --with-libm="$(LIBM)" back). The functional behavior of math/cmath/_statistics is identical to before — same C source, same APIs, same numeric semantics; only the location of libm code changes (now python.elf-only instead of duplicated into each .so).

Enrique Saurez and others added 2 commits June 2, 2026 18:38
[nanvix] E: Phase 1B — build 5 Tier-1 math + memory modules as .so
daa0c44 
Phase 1B of the .a -> .so migration (see
nanvix-todo/cpython-static-to-shared-migration.md section 5).
Builds on Phase 1A (#5) by promoting the remaining 5 Tier-1
"math + memory" stdlib extension modules from statically linked
into python.elf to dlopen-loaded shared objects under
lib/python3.12/lib-dynload/.

Modules moved to *shared* in Modules/Setup.local generation
(.nanvix/docker.py):

- math, cmath, _statistics (libm consumers).
- mmap, _contextvars (libc-only).

None of the five reference external libraries beyond libc / libm
(verified by grepping for #include <zlib|expat|openssl|sqlite|
mpdec|bzlib|lzma|hacl>). libm is already linked into python.elf
via the existing --whole-archive --export-dynamic chain, so the
math / cmath / _statistics .so files resolve sin/cos/etc. at
dlopen time.

Note on .so size: cpython's Makefile attaches libm.a to the link
of math / cmath / _statistics via MODULE_*_LDFLAGS, so the
resulting math.so (~532 KB) and cmath.so (~180 KB) statically
contain a copy of the libm objects they use. This is functionally
correct (--allow-multiple-definition handles the duplicate
symbols at dlopen) but wastes ramfs space. Optimizing the link
to drop libm.a from the .so command line and let the dynamic
loader resolve UND symbols against python.elf is a follow-up
cleanup, not a Phase 1B blocker.

Test coverage (.nanvix/test.py):

- New phase1b_snippet imports each module, asserts it is NOT in
  sys.builtin_module_names, exercises one trivial API call to
  confirm dlopen + PyInit_<name> succeeded, and prints the
  resolved __file__ path. Phase 1A probe retained.

Validation on local toolchain (phase0-llfix):

- All 5 new .so files produced and installed under lib-dynload/
  (sizes: math 532K, cmath 180K, _statistics 28K, mmap 54K,
  _contextvars 11K).
- nm python.elf no longer shows PyInit_<name> for any of the 5.
- python.elf size: 19.31 MB (Phase 1A) -> 19.18 MB (Phase 1B),
  ~130 KB further reduction (smaller than the .so total because
  libm objects are double-counted between python.elf's
  --whole-archive copy and the math / cmath / _statistics .so).
- Hello + Phase 1A probe + Phase 1B probe + lxml + HTTP smoke
  + full regrtest 160/160 PASS in standalone mode.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
[nanvix] E: Drop libm.a from math/cmath/_statistics .so link
7fe6de6 
Phase 1B follow-up. Now that nanvix/nanvix fixes the libm visibility
merge in libposix.a (esaurez/nanvix#26 — localizes the 28 libm wrapper
shadows emitted by Rust's compiler_builtins), python.elf's `.dynsym`
correctly exposes `sqrt`, `cbrt`, `fma`, `ceil`, `fabs`, `fmod`, etc.
as `GLOBAL DEFAULT`. Dlopen'd `.so` modules can now resolve those
symbols at runtime against the main executable, exactly like Linux
dlopen'd modules resolve them against `libm.so.6`.

Pass `--with-libm=` (empty) to CPython's configure so the math /
cmath / _statistics module link commands no longer include libm.a.
Each `.so` ends up with `sqrt` / `cbrt` / etc. as undefined
references that the dynamic loader resolves at dlopen time.

Size impact:
- math.cpython-312.so:        532 KB -> 381 KB  (-28%)
- cmath.cpython-312.so:       180 KB ->  91 KB  (-50%)
- _statistics.cpython-312.so:  28 KB ->  18 KB  (-36%)

python.elf size is unchanged (libm.a was already linked via
--whole-archive into the main binary; that hasn't changed).
Total ramfs reduction: ~240 KB across the three modules.

Validation (on patched toolchain image that bundles
esaurez/nanvix#26's fix to libposix.a / libnvx_crt0.a):
- nm python.elf shows sqrt / cbrt / fma as GLOBAL DEFAULT in .dynsym.
- math.sqrt(4.0)==2.0, cmath.sqrt(-1)==1j round-trip via dlopen.
- Phase 1B import probe: all 5 modules load via dlopen.
- Full CPython regrtest 160/160 modules PASS.
- Hello + lxml + HTTP smoke tests all pass.

Prerequisite: esaurez/nanvix#26 must be merged and the
ghcr.io/nanvix/toolchain-python image rebuilt to carry the patched
libposix.a before this PR can build with the change. Without the
prerequisite, math.so dlopen would fail with "symbol not found" on
sqrt — exactly the failure mode that motivated #26.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
This was referenced Jun 3, 2026
Closed
Closed
esaurez pushed a commit that referenced this pull request Jun 3, 2026
[nanvix] E: Phase 3 — build 7 Tier-3 stdlib extensions as .so
3c60ab2 
Phase 3 of the .a -> .so migration (see
nanvix-todo/cpython-static-to-shared-migration.md section 7).
Promotes 7 modules with external Nanvix-ported .a dependencies
from statically linked into python.elf to dlopen-loaded shared
objects.

Modules moved to *shared* in Modules/Setup.local generation
(.nanvix/docker.py):

- _bz2 (libbz2)        --- unbundled, see Group A below
- _lzma (liblzma)      --- unbundled, see Group A below
- zlib (libz)          --- unbundled, see Group A below
- _sqlite3 (libsqlite3) -- unbundled, see Group A below
- _ssl (libssl + libcrypto) --- still bundled, see Group B below
- _hashlib (libcrypto) --- still bundled, see Group B below
- _ctypes (libffi)     --- still bundled, see Group B below

==============================================================
Group A: 4 sysroot libs unbundled in this PR
==============================================================

For zlib, _bz2, _lzma, _sqlite3 (and binascii which uses libz for
CRC32 via the same ZLIB_LIBS chain), this PR does the full Linux-
style split: the underlying library lives once in python.elf via
--whole-archive --export-dynamic, and each .so resolves its lib
symbols at dlopen time against python.elf's .dynsym.

Same architectural pattern as Phase 1B-drop-libm (#7) established
for libm. Changes in Makefile.nanvix CONFIGURE_ENV:

1. python.elf's LIBS is extended with a --whole-archive ...
   --no-whole-archive group containing libsqlite3.a, libz.a,
   libbz2.a, liblzma.a. The four libs now live exactly once in
   python.elf with every public symbol exported via
   --export-dynamic.

2. The per-module LIBS env vars used by cpython's configure to
   inject -l<lib> into individual .so links are cleared:
   ZLIB_LIBS, BZIP2_LIBS, LIBLZMA_LIBS, LIBSQLITE3_LIBS. The .so
   files now reference compress2 / BZ2_bzCompress / lzma_code /
   sqlite3_exec etc. as UND symbols; the dynamic loader resolves
   them at dlopen time.

Size impact for Group A:

   zlib.so      236 KB -> 181 KB  (-55 KB)
   _bz2.so      128 KB ->  68 KB  (-60 KB)
   _lzma.so     212 KB -> 120 KB  (-92 KB)
   _sqlite3.so 1448 KB -> 473 KB  (-975 KB, biggest .so win)
   python.elf  8.48 MB -> 9.50 MB (+1.02 MB; the 4 libs now baked
                                   in via --whole-archive)

Net ramfs: ~150 KB savings.  Architectural win: each of these 4
sysroot libs now exists exactly once in the image (canonical Linux
pattern).

==============================================================
Group B: 3 modules with still-bundled libs (deferred)
==============================================================

For _ssl, _hashlib, _ctypes, this PR ships the same v1 approach
that Phase 2 used: each .so embeds its own copy of the underlying
library via the existing cpython per-module LDFLAGS mechanism.
The .so files are larger than ideal but functionally complete:

- _ssl.so: 6398 KB (libssl + libcrypto baked in)
- _hashlib.so: 4795 KB (libcrypto baked in; duplicates _ssl's copy)
- _ctypes.so: 744 KB (libffi baked in)

These three are blocked by pre-existing Nanvix sysroot bugs that
prevent --whole-archive wrapping:

1. libcrypto.a (used by _ssl, _hashlib) contains bss_log.o which
   references unimplemented POSIX openlog / syslog / closelog
   (Nanvix's newlib + libposix do not implement these — Nanvix
   logs through a different syslog kcall, not the POSIX client
   API). Under normal archive selection nothing references
   BIO_s_log so bss_log.o is never pulled; under --whole-archive
   every member gets pulled and the link fails. Fix needed:
   either strip bss_log.o from libcrypto.a (post-build), provide
   stub openlog/syslog/closelog, or recompile libcrypto without
   syslog support.

2. libffi.a (used by _ctypes) contains nested archives
   (libposix.a, libc.a, libm.a as ar members). ld --whole-archive
   tries to include every member as an object and fails on the
   nested .a's. Fix needed: either fix the Nanvix libffi build
   (probably a libtool convenience-library mishap) or
   post-process libffi.a to delete the nested members.

Group B and the cpython-internal-libs Group C (libmpdec, libexpat,
libHacl_Hash_SHA2 for _decimal, pyexpat, _sha2) are tracked in
nanvix-todo/phase2-3-unbundle-bundled-libs.md.

==============================================================
Correctness vs the libm case
==============================================================

Unlike the libm case that required nanvix#26 (Rust
compiler_builtins shadows libm's WEAK HIDDEN symbols, breaking
visibility merge), the Phase 3 bundled-lib duplication is a pure
size issue:

- All four Group A libs (libz, libbz2, liblzma, libsqlite3) and
  the three Group B libs (libssl, libcrypto, libffi) are pure C
  with no Rust compiler_builtins shadows.
- None carry shared mutable per-process state in the way libm
  does (FP environment, signgam). The Group B libs DO carry init
  state (OpenSSL algorithm registry, libffi closure caches) but
  each .so copy initializes its own copy — wasteful but not
  incorrect.

For OpenSSL specifically, each .so copy initializes its own
OpenSSL state. Observable side-effect today:
_hashlib.openssl_sha256() called BEFORE any other OpenSSL init
returns "unsupported hash type sha256"; calls from regrtest's
test_hashlib succeed because OpenSSL has been initialized via
_ssl by then. The deferred Group B unbundling fixes this for
free by ensuring exactly one libcrypto init runs in python.elf
before any module dlopens.

==============================================================
Validation
==============================================================

Tested on local toolchain (phase0-llfix overlay containing the
patched libposix.a from esaurez/nanvix#26):

- All 7 .so files produced and installed under lib-dynload/.
- nm python.elf no longer shows PyInit_<name> for any of the 7.
- python.elf size: 16.67 MB (Phase 2) -> 9.50 MB (Phase 3),
  -7.17 MB. Largest single-phase reduction so far.
- Hello + Phase 1A + 1B + 1C + 2 + 3 import probes + lxml + HTTP
  smoke + full regrtest 160/160 PASS in standalone mode.
- test_hashlib, test_ssl, test_zlib, test_bz2, test_lzma,
  test_sqlite3, test_ctypes all included in regrtest's 160 and
  pass.

Phase 1 + Phase 2 + Phase 3 cumulative: 47 of 47 Tier-1/2/3
modules moved from static to .so. python.elf has dropped from
~20 MB at Phase 0 baseline to 9.50 MB now.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@esaurez esaurez mentioned this pull request Jun 3, 2026
Open
@esaurez esaurez force-pushed the feat/phase1b-tier1-mathmem-shared branch from daa0c44 to b91f9d8 Compare June 3, 2026 22:39
@esaurez

esaurez commented Jun 3, 2026

Copy link
Copy Markdown
Owner Author

Folded into PR #6 (Phase 1B) per esaurez review preference — easier to review the .so move and the libm-unbundling together as one cohesive change. The squashed Phase 1B commit on eat/phase1b-tier1-mathmem-shared now contains both.

@esaurez esaurez closed this Jun 3, 2026
@esaurez esaurez deleted the feat/phase1b-drop-libm-from-math-so branch June 3, 2026 22:39
esaurez pushed a commit that referenced this pull request Jun 3, 2026
[nanvix] E: Phase 1C — build 8 Tier-1 text-codec modules as .so
0866b55 
Phase 1C of the .a -> .so migration (see
nanvix-todo/cpython-static-to-shared-migration.md section 5).
Builds on Phase 1B (#6, #7) by promoting the remaining 8 Tier-1
"text codec" stdlib extension modules from statically linked into
python.elf to dlopen-loaded shared objects under
lib/python3.12/lib-dynload/.

Modules moved to *shared* in Modules/Setup.local generation
(.nanvix/docker.py):

- unicodedata: Unicode database lookups (the big one — 1.2 MB of
  unicode data tables).
- _multibytecodec: shared CJK codec infrastructure.
- _codecs_cn / _codecs_hk / _codecs_iso2022 / _codecs_jp /
  _codecs_kr / _codecs_tw: per-region CJK codec tables.

None of the eight reference external libraries; they are pure C
with embedded data tables. They link against the same -lc /
runtime symbols that the rest of the Phase 1 modules use.

Test coverage (.nanvix/test.py):

- New phase1c_snippet imports each module, asserts it is NOT in
  sys.builtin_module_names, exercises one trivial API call to
  confirm dlopen + PyInit_<name> succeeded (unicodedata.lookup,
  _multibytecodec.__create_codec, _codecs_<region>.getcodec), and
  prints the resolved __file__ path. Phase 1A/1B probes retained.

Validation on local toolchain (phase0-llfix):

- All 8 new .so files produced and installed under lib-dynload/
  (unicodedata 1193K, _codecs_jp 262K, _codecs_hk 168K,
  _codecs_cn 155K, _codecs_kr 145K, _multibytecodec 147K,
  _codecs_tw 115K, _codecs_iso2022 76K — total ~2.2 MB across
  the eight files).
- nm python.elf no longer shows PyInit_<name> for any of the 8.
- python.elf size: 19.18 MB (Phase 1B) -> 17.48 MB (Phase 1C),
  -1.70 MB. Biggest single-phase reduction so far because the
  CJK codec tables and the Unicode database are large.
- Hello + Phase 1A + Phase 1B + Phase 1C import probes + lxml +
  HTTP smoke + full regrtest 160/160 PASS in standalone mode.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
esaurez pushed a commit that referenced this pull request Jun 3, 2026
[nanvix] E: Phase 3 — build 7 Tier-3 stdlib extensions as .so
81de428 
Phase 3 of the .a -> .so migration (see
nanvix-todo/cpython-static-to-shared-migration.md section 7).
Promotes 7 modules with external Nanvix-ported .a dependencies
from statically linked into python.elf to dlopen-loaded shared
objects.

Modules moved to *shared* in Modules/Setup.local generation
(.nanvix/docker.py):

- _bz2 (libbz2)        --- unbundled, see Group A below
- _lzma (liblzma)      --- unbundled, see Group A below
- zlib (libz)          --- unbundled, see Group A below
- _sqlite3 (libsqlite3) -- unbundled, see Group A below
- _ssl (libssl + libcrypto) --- still bundled, see Group B below
- _hashlib (libcrypto) --- still bundled, see Group B below
- _ctypes (libffi)     --- still bundled, see Group B below

==============================================================
Group A: 4 sysroot libs unbundled in this PR
==============================================================

For zlib, _bz2, _lzma, _sqlite3 (and binascii which uses libz for
CRC32 via the same ZLIB_LIBS chain), this PR does the full Linux-
style split: the underlying library lives once in python.elf via
--whole-archive --export-dynamic, and each .so resolves its lib
symbols at dlopen time against python.elf's .dynsym.

Same architectural pattern as Phase 1B-drop-libm (#7) established
for libm. Changes in Makefile.nanvix CONFIGURE_ENV:

1. python.elf's LIBS is extended with a --whole-archive ...
   --no-whole-archive group containing libsqlite3.a, libz.a,
   libbz2.a, liblzma.a. The four libs now live exactly once in
   python.elf with every public symbol exported via
   --export-dynamic.

2. The per-module LIBS env vars used by cpython's configure to
   inject -l<lib> into individual .so links are cleared:
   ZLIB_LIBS, BZIP2_LIBS, LIBLZMA_LIBS, LIBSQLITE3_LIBS. The .so
   files now reference compress2 / BZ2_bzCompress / lzma_code /
   sqlite3_exec etc. as UND symbols; the dynamic loader resolves
   them at dlopen time.

Size impact for Group A:

   zlib.so      236 KB -> 181 KB  (-55 KB)
   _bz2.so      128 KB ->  68 KB  (-60 KB)
   _lzma.so     212 KB -> 120 KB  (-92 KB)
   _sqlite3.so 1448 KB -> 473 KB  (-975 KB, biggest .so win)
   python.elf  8.48 MB -> 9.50 MB (+1.02 MB; the 4 libs now baked
                                   in via --whole-archive)

Net ramfs: ~150 KB savings.  Architectural win: each of these 4
sysroot libs now exists exactly once in the image (canonical Linux
pattern).

==============================================================
Group B: 3 modules with still-bundled libs (deferred)
==============================================================

For _ssl, _hashlib, _ctypes, this PR ships the same v1 approach
that Phase 2 used: each .so embeds its own copy of the underlying
library via the existing cpython per-module LDFLAGS mechanism.
The .so files are larger than ideal but functionally complete:

- _ssl.so: 6398 KB (libssl + libcrypto baked in)
- _hashlib.so: 4795 KB (libcrypto baked in; duplicates _ssl's copy)
- _ctypes.so: 744 KB (libffi baked in)

These three are blocked by pre-existing Nanvix sysroot bugs that
prevent --whole-archive wrapping:

1. libcrypto.a (used by _ssl, _hashlib) contains bss_log.o which
   references unimplemented POSIX openlog / syslog / closelog
   (Nanvix's newlib + libposix do not implement these — Nanvix
   logs through a different syslog kcall, not the POSIX client
   API). Under normal archive selection nothing references
   BIO_s_log so bss_log.o is never pulled; under --whole-archive
   every member gets pulled and the link fails. Fix needed:
   either strip bss_log.o from libcrypto.a (post-build), provide
   stub openlog/syslog/closelog, or recompile libcrypto without
   syslog support.

2. libffi.a (used by _ctypes) contains nested archives
   (libposix.a, libc.a, libm.a as ar members). ld --whole-archive
   tries to include every member as an object and fails on the
   nested .a's. Fix needed: either fix the Nanvix libffi build
   (probably a libtool convenience-library mishap) or
   post-process libffi.a to delete the nested members.

Group B and the cpython-internal-libs Group C (libmpdec, libexpat,
libHacl_Hash_SHA2 for _decimal, pyexpat, _sha2) are tracked in
nanvix-todo/phase2-3-unbundle-bundled-libs.md.

==============================================================
Correctness vs the libm case
==============================================================

Unlike the libm case that required nanvix#26 (Rust
compiler_builtins shadows libm's WEAK HIDDEN symbols, breaking
visibility merge), the Phase 3 bundled-lib duplication is a pure
size issue:

- All four Group A libs (libz, libbz2, liblzma, libsqlite3) and
  the three Group B libs (libssl, libcrypto, libffi) are pure C
  with no Rust compiler_builtins shadows.
- None carry shared mutable per-process state in the way libm
  does (FP environment, signgam). The Group B libs DO carry init
  state (OpenSSL algorithm registry, libffi closure caches) but
  each .so copy initializes its own copy — wasteful but not
  incorrect.

For OpenSSL specifically, each .so copy initializes its own
OpenSSL state. Observable side-effect today:
_hashlib.openssl_sha256() called BEFORE any other OpenSSL init
returns "unsupported hash type sha256"; calls from regrtest's
test_hashlib succeed because OpenSSL has been initialized via
_ssl by then. The deferred Group B unbundling fixes this for
free by ensuring exactly one libcrypto init runs in python.elf
before any module dlopens.

==============================================================
Validation
==============================================================

Tested on local toolchain (phase0-llfix overlay containing the
patched libposix.a from esaurez/nanvix#26):

- All 7 .so files produced and installed under lib-dynload/.
- nm python.elf no longer shows PyInit_<name> for any of the 7.
- python.elf size: 16.67 MB (Phase 2) -> 9.50 MB (Phase 3),
  -7.17 MB. Largest single-phase reduction so far.
- Hello + Phase 1A + 1B + 1C + 2 + 3 import probes + lxml + HTTP
  smoke + full regrtest 160/160 PASS in standalone mode.
- test_hashlib, test_ssl, test_zlib, test_bz2, test_lzma,
  test_sqlite3, test_ctypes all included in regrtest's 160 and
  pass.

Phase 1 + Phase 2 + Phase 3 cumulative: 47 of 47 Tier-1/2/3
modules moved from static to .so. python.elf has dropped from
~20 MB at Phase 0 baseline to 9.50 MB now.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
This was referenced Jun 3, 2026
Open
Open
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@esaurez