Fix interface with xgap and GAP.app (WindowCmd corrupting long answers or ones containing @)

[fingolfin]

SyWinCmd read the answer @a<len>+<data> of a window handler into a static buffer of 8000 bytes, which a longer answer overflowed. An answer arriving in several read() calls was corrupted, as the read loop did not advance its destination, and a failed read or an end of file made it spin. The un-escaping pass counted the two bytes of an escape as one and so ran past the end of the answer.

SyWinCmd now reads exactly len bytes into a string object, retrying on EINTR and EAGAIN, and un-escapes them in place. FuncWindowCmd checks bounds while parsing: a string entry exceeding the answer, a number not fitting a small integer and an unknown entry raise an error. The answer is consumed before it is parsed, so these errors leave the input in sync.

In package mode the answer is read from stdin, where GAP does not read ahead, so a testspecial test can put it on the line after the call. For this run_gap.sh passes on the GAP options a test asks for in a first line #GAPOPTS <options>. The answers carry several interdependent lengths, hence the tests are generated by dev/make-window-cmd-tests.g.

This is based on the analysis and a first fix by @RussWoodroofe.

Assisted-by: Claude Code (Fable 5.1)

---

Alternative to and hence closes #6483. Tested with xgap; that also lead to gap-packages/xgap#38. Not tested via GAP.app mainly because I don't know how (and its SourceForget git repository was last updated 2022?)

The main motivation for this PR was a desire to have a somewhat simpler kernel implementation, and also a tighter test rig.

[codecov]

Codecov Report

❌ Patch coverage is 84.81013% with 12 lines in your changes missing coverage. Please review.
✅ Project coverage is 79.62%. Comparing base (a872f30) to head (1f7d325).
⚠️ Report is 56 commits behind head on master.

Files with missing lines	Patch %	Lines
src/sysfiles.c	77.55%	8 Missing and 3 partials ⚠️
src/gap.c	96.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #6584      +/-   ##
==========================================
+ Coverage   79.03%   79.62%   +0.58%     
==========================================
  Files         683      683              
  Lines      295287   307167   +11880     
  Branches     8642     9519     +877     
==========================================
+ Hits       233375   244574   +11199     
- Misses      60096    60685     +589     
- Partials     1816     1908      +92     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[RussWoodroofe]

I looked through your code. If I understand right, a main difference with what I had is that you read the response directly into a GAP string object, and unescape it there. That probably is a little better design. I didn't look in super detail, but the design is sound, and it is easier to compare with prior version. I'm not an expert on GAP memory management, so my check on that is not reliable.
You did also dropped the timeout in the test system -- I guess that on the automated github system, it will terminate it anyway?

Testing: I applied the patch to a 4.16.1 tree and ran with Gap.app, both upcoming 0.8 release and also 0.7a. I didn't see any trouble with either, including with long answers to WcDialog. I will try to use this as my daily driver, and will report issues if I see anything.

Sorry to take a day or so to get back here.

[RussWoodroofe]

One other comment: the test framework that is here may tend to mostly obsolete the package mode testing framework that I contributed to xgap some time ago.