ponder-lab / Hybridize-Functions-Refactoring / #3235
87%
main: 88%

LAST BUILD BRANCH: 9ad4fea94e0e87f5a5887fdaa82ed29d38add636
DEFAULT BRANCH: main
Ran
Jobs 1
Files 44
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 86.638% (+0.02%) from 86.617%
#3235

push

github

web-flow 
Re-infer a developer's stripped input signature and score the two (#993)

* Re-infer a developer's stripped input signature and score the two

Add a strip-and-re-infer harness under edu.cuny.hunter.hybridize.eval/reinfer.
For each fix commit in a manifest, it strips only the input_signature argument
of the named tf.function decorators, runs the evaluator annotation-free with
inference on, and relates the removed spec to the re-inferred one per function,
per parameter and per axis.

The strip cuts the keyword by its exact source span and verifies the edit on
the AST: the original module with that one keyword deleted must equal the
edited module. The relation restates InputSignature.relate, plus the same
order restricted to dtype and to shape. relation_cases.txt holds the cases
both the harness's tests and InputSignatureTest read, so the two orders cannot
drift. Every function gets one outcome (scored, not reproduced with a reason,
no call site, or excluded), and the call-site evidence behind "no call site"
is kept as columns so it can be audited.

CI runs the harness's unit tests after Black.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NTnpjPjJj2aZy1ZkUD1eAn

* Record whether a run's harness is merged

The commit alone cannot answer that after a squash merge, which gives the same
code a new commit. The run record now also carries the git tree of the harness
directory, the tree of that directory on main, and whether the commit is an
ancestor of main, so a run can be shown to have used merged code either way.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NTnpjPjJj2aZy1ZkUD1eAn

* Keep a failed evaluation's functions in the output

A subject whose evaluation does not complete, for example because the analysis
exhausts the heap, now contributes a row per function with the outcome
evaluation-failed:<cause>, instead of dropping out of the... (continued)

4085 of 4715 relevant lines covered (86.64%)

0.87 hits per line

Jobs
ID Job ID Ran Files Coverage
1 #3235.1 44
86.64
Source Files on build #3235