Regenerate all the checksums

[ViralBShah]

I ran the entire checksum script (mainly to pick up the libblastrampoline update) and noted a bunch of other things the script modified. I figure it is best to update it wholesale with the output of the script on all the deps.

It also created deps/checksums/llvmunwind, which was not already checked in and is added in here.

MPFR in Versions.make was out of sync with MPFR_jll in the stdlibs.

[giordano]

It also created deps/checksums/llvmunwind, which was not already checked in and is added in here.

There is an empty deps/checksums/llvmunwind and the llvm libunwind checksums were added to deps/checksums/unwind. Is that right?

[ViralBShah]

I didn't pay too much attention to that. I just assumed the auto-generate script will do the right thing. It is possible the script needs updating. Maybe the unwind target needs some cleanup.

[DilumAluthge]

@ViralBShah Can you post the exact make invocations that you ran?

@vtjnash It sounds like the refresh_checksums script is not packing the libunwind checksums into the correct file?

[ViralBShah]

On a clean clone of julia, I ran

make -f contrib/refresh_checksums.mk  -j

[ViralBShah]

deps/unwind.mk is where all the llvmunwind stuff is. Perhaps that file should be called deps/llvmunwind.mk?

[DilumAluthge]

Maybe the refresh_checksums.mk script doesn't work when run in parallel? Can you try in a fresh clone of Julia without the -j?

If that fixes the problem, then we should add .NOTPARALLEL: to the top of that script.

[ViralBShah]

Ok, how does this look now? I definitely think the script needs a .NOTPARALLEL. Would be nice to regenerate these automatically, but it takes a long time to run. Maybe we can do it every once in a while.

[DilumAluthge]

It's still creating deps/checksums/llvmunwind, but I think those checksums should be going into deps/checksums/unwind alongside the Nongnu Unwind checksums. That is, I think we want to store all the unwind library checksums in the same file.

[DilumAluthge]

@vtjnash Is there an extra command that Viral needs to run to pack all the unwind checksums into a single file?

[vtjnash]

That sounds odd. They should definitely be ending up in different files by following the rules (even in parallel). I don't know how that got out of sync, but once out of sync, it can remain a bit confused.

[vtjnash]

Look reasonable to me. The goal in the future is that any PR updating one of these files should likely be updating every single line in the file. That makes confirmation easy and minimizes cross-PR contamination.