* fix(artifacts): verify FAT identities before Bun and Node SEA claims
* fix(node-sea): bound thin Mach-O header reads to their slice
* fix(bun): exclude rejected FAT slices from trailer fallback
Reuse the admitted JSON comparator for environment replay and desktop task checks. Keep the public missing-operand types and portable Python semantics intact.
The XML plist reader replaced every literal CR and CRLF with LF before
parsing. CoreFoundation keeps them, so REA reported a different string than
plutil for the same file, and two keys that differ only in their line ending
collapsed into one repeated key.
The document is now parsed as written. Source spans for integer, real, date
and data literals already count CR, CRLF and LF as one line break each.
Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
* fix(native): keep an assumed relocation width from reporting a section overrun
inspect_native_initializers assumed that a relocation of a type it does not
size writes a whole word. For a 4-byte field that ends its section, such as
R_X86_64_PLT32 or R_AARCH64_CALL26 on a tail call that ends .text, the assumed
word ran past the section, so the object was reported incomplete with
"A section relocation extends beyond its target mapping".
An assumed width is now narrowed to the bytes its mapping holds. A relocation
of known width that runs past its section is still reported.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
* test(native): use the tail-call relocation clang emits for AArch64
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
* fix(native): size link-time relocation kinds instead of narrowing an assumed width
Review found that narrowing an assumed word to its mapping hid a real overrun
of an unlisted kind and, in a linked image, a slot behind zero-fill bytes.
The mapping remainder does not establish what a relocation writes.
The section walk is restored to its previous logic. The width table now sizes
the link-time kinds clang emits into x86-64 and AArch64 objects, as the i386
row already does, so a call that ends .text is a known 4-byte field. A kind
the table does not hold keeps its assumed word and its overrun note.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
* test(native): cover a newly sized kind in the loader relocation walk
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
* fix(ci): count named artifact transfers in job timings
* fix(ci): cover named artifact timings against their workflows
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(app-bundle): apply the plist decode budget to the XML branch
* test(app-bundle): cover the XML plist decode budget
* decode xml plist once, pass the text to the decode preflight
* fix(app-bundle): keep OpenStep plists out of the XML preflight, keep budget failures typed
OpenStep Info.plist text is not XML, so quoted markup inside it is string
data. Run the XML entity-expansion preflight only on actual XML input,
using the same format check the parser already uses. A valid OpenStep
bundle with a DOCTYPE-looking quoted string no longer fails the budget.
Budget exhaustion also keeps its own error: the resolver now returns a
typed bundle_info_plist_decode_budget_exceeded failure naming the 16 MiB
allowance instead of the malformed-plist diagnosis.
* Fix test names and budget error classification after the merge
* fix: regenerate user-visible error catalogue for plist budget constraint
* test: pin budget-exceeded bundle target to its catalogue row
* refactor(plist): give Info.plist decoding a shared owner
---------
Co-authored-by: Dextheking1 <Dextheking1@users.noreply.github.com>
Co-authored-by: benjaminkincses <4175097+benjaminkincses@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>