Files
ARCADIA-Studio/harness/experiments/analysis-text-producer-20260925/STACK-OBSERVATIONS.md
T

49 lines
2.7 KiB
Markdown

# Text producer native frame observations
Source `97a8703adc75467a7f8e75f5c6cb00ca542b707c`, Windows gnullvm debug
test artifact 42,614,784 bytes, SHA256
`41da1a4d788e2bd7392e69a724b7de259752d9f813ae6b42eadf580ec682a54a`.
The six new tests pass. The complete 275-name library run finishes with
259 passed and 16 inherited failed, including all ten stack-named controls
passing. No old test or stack allowance was changed. The new text-mode cases
run on the ordinary harness stack; they do not establish a separate 128 KiB
stronger-mode guarantee.
Read-only llvm-nm/llvm-objdump helpers bind their observations to the focused
receipt's exact artifact. Before rebuilding, the same helpers captured the
8d52dab artifact, SHA256
`29d8a4761fd8f09a3c0eafeb11cca07f0d45c711bf1b969d97c8b29d2b3ab446`.
These prior wrapper observations were recorded after the emission evidence ZIP
was frozen; they belong to this packet, not its predecessor's 65 members.
| Selected prologue reservation | 8d52dab | 97a8703 | Change |
| --- | ---: | ---: | ---: |
| parse_impl | 11,048 | 11,560 | +512 |
| Parser::unit | 11,864 | 12,072 | +208 |
| accepted test helper | 1,432 | 1,528 | +96 |
| indexed test helper | 1,496 | 1,560 | +64 |
| ordinary-index worker closure | 6,920 | 7,176 | +256 |
These selected reservations increase by 1,136 bytes in total. The 56-byte
same-named test shim is distinct from the actual worker. parse_with_profile
(200), parse (280), State::parse_typed (6,952), declarations::parse (22,664),
binding (2,520), prefix (11,720), continuation (13,416), Exact::parse_decimal
(2,296), parse_literal (952), Budget::exact (632) and Cursor::exact_literal
(776) remain unchanged in the captured prologues. Prefix and continuation are
mutually exclusive and must not be summed as simultaneously active helpers.
The initial6b0ef99 artifact (42,615,808 bytes, SHA256
`d3f8f24775f81fd822c01ca887524de9bd80b06de8b3bff8dff550cd3b8d98a6`)
has its own preserved observations. Every selected reservation in that candidate
matches the final97a8703 one. Its strict Clippy failure was corrected by direct
if-let matching; initial passing runtime results are not substituted for the
final rebuilt source's separately recorded tests.
This is a finite compiler-output comparison, not a complete measured live stack,
remaining headroom or a guarantee across inputs/platforms/process layouts. It
retains the earlier parser split while exposing the cost of added scalar Budget
state. The old overflow control passes in this actual full run; the predecessor's
three extra process rechecks are not relabeled as runs of this source. No new
debugger session, stack override, allocator wrapper or retry-to-pass was used.
The separate historical full-pipeline setup observation remains unrerun.