[Bug manual/28956] New: Name collision for RISC-V in libm ULP error table

[thomas.coldrick at codethink dot co.uk]

https://sourceware.org/bugzilla/show_bug.cgi?id=28956

            Bug ID: 28956
           Summary: Name collision for RISC-V in libm ULP error table
           Product: glibc
           Version: unspecified
            Status: UNCONFIRMED
          Severity: minor
          Priority: P2
         Component: manual
          Assignee: unassigned at sourceware dot org
          Reporter: thomas.coldrick at codethink dot co.uk
                CC: mtk.manpages at gmail dot com
  Target Milestone: ---

As reported in this mailing list post [0], the libm-test-ulps-name file for the
32- and 64-bit variants of RISC-V both contain "RISC-V", which means that the
generated manual can contain either the 32- or 64-bit variant with no
indication as to which it is. The differences are minor, but this could mislead
users of the documentation who are after the specific value for their
architecture.

In addition to misleading, this also has the effect of making the build of
glibc non-reproducible, as the variant selected is dependent upon the ordering
of a python os.walk() call, which varies based on the filesystem.

In the mailing list thread, it was questioned whether there's much value in
tracking both RISC-V variants separately. I lack the domain knowledge to
comment comprehensively on this (I found this bug due to build reproducibility
tests failing), but the fact that the variants differ in some mathematical
functions' error values suggests to me that there is probably at least some
value in this.

To fix this, I propose that we rename the names of the variants to "RISC-V
32-bit" and "RISC-V 64-bit" respectively in the libm-test-ulps-name files.

-- 
You are receiving this mail because:
You are on the CC list for the bug.