GeezerVison

Copy link

Copy Markdown

Member

yan12125, alindt, mszczepanczyk, and Sipty reacted with thumbs up emoji

zhangyangyu force-pushed the workspace branch from 8384aca to 601b3f7 Compare February 16, 2017 05:31

…tor and traceback objects

__setstate__ must accept the state returned by __reduce__. This was not the case for generator
and trace-back objects. This commit fixes this. The next commit (merge of issue python#127) adds the relevant test cases.
Additionally amends changelog.txt.

https://bitbucket.org/stackless-dev/stackless/issues/107

…ling tests.

This change is based on f5f98595c6cc63b from 2.7-slp, but it is heavily modified for Python 3.

- Pickle 'callable-iterator' objects correctly. Previously the unpickled object had the type '_stackless._wrap.callable-iterator'.

- Fix pickling of 'method-wrapper' objects. Previously pickling them caused a SystemError exception.

- Fix copy.copy() for 'callable-iterator', 'method', 'dict_keys', 'dict_values' and 'dict_items' objects. Previously the copied object had the type '_stackless._wrap....'.

- Fix Stackless pickling tests. The method StacklessTestCase.dumps() didn't pass the pickle protocol to the pickler.

- Remove dead code in prickelpit.c. The code was used in older Stackless versions.

…iterator'

Disable the Stackless specific code for pickling 'iterator' and 'callable_iterator' objects. C-Python 3.3 already pickles them.
Add tests to ensure, that Stackless can still unpickle old pickles.

https://bitbucket.org/stackless-dev/stackless/issues/127

raspy reviewed Apr 17, 2018

View reviewed changes

auvipy reviewed Oct 9, 2018

View reviewed changes

Sorry, something went wrong.

Uh oh!

There was an error while loading. Please reload this page.

serhiy-storchaka reviewed Oct 29, 2018

View reviewed changes

Sorry, something went wrong.

Uh oh!

There was an error while loading. Please reload this page.

Pays the structural debt logged by W-C2 push 45 (9e342ab). The
W-C2 fix overloaded is_abs_addr=1 to trigger SIB pure-disp32 encoding
in phx_fs_ptr because PhxMem zero-init couldn't distinguish 'no base
register' from 'base = RAX (id=0)'. That overload was a workaround,
not a fix to the underlying flaw — pythia python#127 flagged the
structural-fix-rejected pattern.

Fix: add explicit has_base flag to PhxMem (mirror of has_index).
Constructors that take a base register (phx_ptr, phx_ptr_index,
phx_a64_mem/_pre/_post) set has_base=1; pure-displacement constructors
(phx_fs_ptr) leave has_base=0 via zero-init. encode_modrm_mem branches
on !has_base between the is_abs_addr branch and the base-register
fallback, emitting the SIB pure-disp32 form (04 25 disp32). FS segment
prefix is independently emitted by emit_segment_prefix.

phx_fs_ptr no longer needs the is_abs_addr workaround; restored to
clean semantics (has_base=0 from zero-init is the structural marker).

Other PhxMem callsite audit (no other changes needed):
- frame_asm_c.c:198 label_mem: is_label_rel branch fires first
- autogen_translate_c.c:1649 x86_operand_to_mem: STACK/IND use phx_ptr/
  phx_ptr_index (has_base=1); MEM uses is_abs_addr (branch fires first)
- arch.c:45+55 jit_arch_ptr_offset/_resolve: returns from phx_ptr
  (has_base=1)

Note (librarian 17:27:04Z): the 1afc0fa 2026-04-06 commit was a
stale-reference cleanup of a never-existing field, not a deliberate
has_base rejection. DSecondary-2 is fresh ground; no prior decision is
being reversed.

Auth chain: theologian patch-shape APPROVE 17:50:43Z (structural
verification: encoder branch order is_abs_addr → is_label_rel →
!has_base → base-register fallback; ARM64 helpers correctly set
has_base=1; other-callsite audit consistent); testkeeper pre-commit
BUILD PASS + JIT smoke PASS 17:53:26Z; supervisor 17:18:11Z queue
re-sequence (pay structural debt while surface fresh).