Carbon Explorer Fails with Error CHECK in NumPossibleCPUsNoCache: cpus.has_value() (false)

[axsaucedo]

Description of the bug:

Carbon Explorer fails with error NumPossibleCPUsNoCache: cpus.has_value() (false)

Full error:

1 external/tcmalloc~/tcmalloc/internal/sysinfo.cc:124] CHECK in NumPossibleCPUsNoCache: cpus.has_value() (false) 
Program terminated with signal: SIGSEGV
Compiler returned: 139

Screenshot

What did you do, or what's a simple way to reproduce the bug?

Opened the default example in Chrome / Firefox and error appeared.

What did you expect to happen?

Default examples should work.

What actually happened?

Errors above.

Any other information, logs, or outputs that you want to share?

No response

[jonmeow]

@chandlerc It looks like this is just tcmalloc not handling compiler-explorer's environment. I expect having that work is more important than having tcmalloc, so I'll send you a rollback.

[jonmeow]

Thanks for reporting this, I've sent out a change to disable it. Note it'll probably be a day for compiler-explorer to pick up a new build.

[jonmeow]

Filed google/tcmalloc#245 regarding this

[jonmeow]

tcmalloc is probably going to keep the behavior. Following up with compiler-explorer to see about whitelisting the /sys file now: compiler-explorer/compiler-explorer#6734

[jonmeow]

Note, this also follows some discussion with @zygoloid that even though glibc has a fallback for this situation, we will probably only support linux versions that have the /sys file.

[axsaucedo]

Thank you for the prompt reply and context @jonmeow - it seems the carbon explorer is working for the time being with the workaround

[chandlerc]

Root cause should be fixed in Compiler Explorer now: compiler-explorer/compiler-explorer#6734

Sent a PR to re-enable TCMalloc and once it's live and not crashing, should be able to resolve this fully.