Return a failure instead of crashing if shape inference can not be run because of unraked operand types

[jorickert]

Some computeShape functions assume/rely on the operands being ranked.
This PR adds a check to them, so that they return failure in case this assumption does not hold, instead of asserting. This allows callers of this function to handle this failure

[jenkins-droid]

Can one of the admins verify this patch?

[AlexandreEichenberger]

@jenkins-droid test this please

[AlexandreEichenberger]

LGTM,

I verified a few, let me just ask just to be sure. The return failure() that are added here are all in the computeShape(...) functions, which is fine since we expect to have successfully run shape inference before, were potentially unranked values have been at least ranked.

Thanks for adding "graceful" errors, much appreciated.

[jorickert]

@AlexandreEichenberger Yes, the return failure() are all added in computeShape . This allows callers of this function to decide how to handle failures in case computeShape fails because the input is unranked.

Before this PR, the functions computeShapeAndAssertOnFailure and computeShape would behave the same in many cases, as computeShape would assert internally.

It would be nice if you could merge this, as I do not have merge rights

[AlexandreEichenberger]

Will do, for some reason I don't have the usual button to update to the latest. do you mind doing this on your end, and then I will merge. I asked for you to have access right to run the CIs. After a PR or two, will give you merge permission too.

[jorickert]

@AlexandreEichenberger I rebased onto the latest main

[jenkins-droid]

Jenkins Linux ppc64le Build #15093 [push] Return a failure instead... started at 11:38

[jenkins-droid]

Jenkins Linux s390x Build #16066 [push] Return a failure instead... started at 11:22

[jenkins-droid]

Jenkins Linux amd64 Build #16064 [push] Return a failure instead... started at 10:22

[jenkins-droid]

Jenkins Linux amd64 Build #16064 [push] Return a failure instead... failed after 53 min

[jenkins-droid]

Jenkins Linux s390x Build #16066 [push] Return a failure instead... passed after 1 hr 27 min

[jorickert]

Jenkins Linux amd64 Build #16064 [push] Return a failure instead... failed after 53 min

@AlexandreEichenberger
The CI error looks like some issue with the CI setup itself to me:
s = '[0.002s][warning][perf,memops] Cannot use file /tmp/hsperfdata_root/232111 because it is locked by another process (e...QAAAAABAAAAAQEAAAABAQEBAAABAAABAQABAAEAAAEBAAEBAQABAQABAAAAAQEBAQEAAAABAQEBAQEAAQAAAAAAAQEAAAEAAAABAQAAAQEAAQEBAAEA"}]'

[jenkins-droid]

Jenkins Linux ppc64le Build #15093 [push] Return a failure instead... passed after 2 hr 18 min