CI hangs in some timeout. Probably not related to MR. Can you re trigger CI run?

I don't think I can...

No idea what the issues is.
After hours of getting it to build on windows, it doesn't run into a timeout.
Since it also is the same code as before #810, I assume whatever the issue is was there before...

I could exclude the case from the build matrix 🤷

Let me take a stab at it tonight.

See the failures in #823

It's due to quickjs-ng/quickjs@7304c3d which means we have some mismatched allocation thing going on, and likely what breaks the test is an underflow.

I'm digging into it.

Forgot to leave this here, since it ended up being late yesterday 😅

The 1st step to figuring it out was to split the ffi tests into individual files. I have done this several times already since it really helps narrow down failures.

Then I thought I'd see if Claude could figure it out before busting out my Windows machine. Right on the money:

On Windows x86_64 (LLP64 data model), sizeof(long) = 4, but ffi_call requires the return buffer to be at least sizeof(ffi_arg)
  bytes (which is 8 on x86_64). When the return type is sint or float (4 bytes), the buffer is only 4 bytes, but ffi_call writes 8
   bytes into it — a heap buffer overflow.

  - Debug mode catches it: MSVC debug CRT adds guard bytes around allocations. The overflow corrupts them, and on free the CRT
  triggers _ASSERTE which shows a modal dialog box. In CI/headless, the dialog blocks forever → timeout.
  - mimalloc masks it: no debug guard bytes, so the overflow goes undetected.
  - Release mode gets lucky: the standard CRT doesn't check for overflows.

  The two failing tests both call C functions returning int (4 bytes). Tests returning pointers/structs/size_t/int64 (all >= 8
  bytes) allocate enough space and pass fine.

That explains the timeout, and all is good now!

Ah thanks for the explanation. I wonder why it didn't occur on my win VM.
But who knows it was an awful experience getting it to build there at all.
E.g. had to reduce src/cacert.h to even get it build, compiler always complained about not enough heap (on a 24GB RAM VM). I mean how can you blow up your AST and symbols to have issues with a 250Kb source file 🤦‍♂️

Anyway thanks for the work!