libera/#clasp - IRC Chatlog

12:03:32 yitzi drmeister: I've added the cando regression tests to the testing workflow on cando-developers/cando. You can run the cando regression tests with `ninja -C build cando-test`

13:00:30 yitzi Bike: #1349 is ready for review. Put a some explanation of the logic in the PR comments.

14:57:14 drmeister Hello

14:58:17 yitzi Hi!

17:15:02 yitzi Bike: I think the logical pathname are getting mangled in `llvm-sys:create-file` and the usage of `llvm::DIFile`. The logic there seems to assume physical paths with `/` separators.

17:18:32 Bike the logic in DIBuilder::createFile?

17:20:16 yitzi Meant the usage of createFile. Specifically make-file-metadata.

17:22:05 yitzi https://github.com/clasp-developers/clasp/blob/aba76e80d3074243241031829ac0dd5486ffd50d/src/lisp/kernel/cmp/debuginfo.lisp#L110-L123

17:22:07 Bike ahh, i see. good, that's a lot more fixable

17:23:09 yitzi I am making an attempt at it. Basically, not using the directory field at all and just shoving the whole namestring into the Filename

17:24:36 yitzi Found that while using STEP the returned pathnames for the frame stack was things like `/src/lisp/kernel/....` vs `sys:src;lisp;kernel;...`

17:26:11 Bike maybe we could check if the pathname is logical (i think (typep pathname 'logical-pathname)?) and choose a separator appropriately, if sticking everything in the file field doesn't work

17:27:03 yitzi Yes, that would make sense. The only obstacle might be if a physical pathname has a host or a device component?

17:27:05 Bike oh, or actual directory-namestring itself screws it up

17:27:20 Bike (directory-namestring #P"SYS:SRC;LLVMO;DEBUGINFOEXPOSE.CC") => "/SRC/LLVMO/"

17:27:35 yitzi Yeah....not good.

17:28:33 Bike maybe we can just change that. it looks like on sbcl directory-namestring on a logical pathname keeps the ; separators.

17:28:48 Bike (checking because pathnames are an area of the standard i'm less familiar with)

17:29:14 yitzi I am changing the corresponding `getPath` function so that it checks for an empty directory string. `(directory.empty()) ? filename : (directory + "/" + filename);`

17:30:15 yitzi Hopefully that will work if some other code creates `llvm::DIFile` and actually uses the directory slot.

17:30:55 yitzi Does SBCL keep the host is directory-namestring?

17:31:02 yitzi s/is/in/

17:31:02 Bike no.

17:31:45 yitzi That would make it difficult. We wouldn't know what logical host the DIFile is pointing to.

17:32:35 Bike maybe we could use enough-namestring instead of directory-namestring?

17:33:13 Bike except i guess that kind of works in the opposite direction

17:33:49 yitzi Well, maybe not using the Directory slot will work. Fingers crossed.

17:47:49 yitzi That seems to work, although something is still shoving a `/` in front.