Re: [patch, gas, documentation] Some additional testsuite info

[Jan Beulich <jbeulich@suse.com>]

On 30.04.2023 12:16, Thomas Koenig via Binutils wrote:
> here's some additional info on the gas testsuite.  I also took the
> liberty to delete the "kind of lame" comment, which is not really
> appropriate any more, at least not for the architectures which
> were added in the last few decades.

Up-front remark: Non-inlined patches are hard to comment upon.

You refer to "the main binutils directory", which is ambiguous next
to "the main gas directory". Itym the top-level directory there
(and not e.g. the binutils/ one being a sibling of gas/).

The reference to "a @code{nop} instruction" is ambiguous, too, I'm
afraid: There's no such requirement (anymore) if you mean to refer
to an instruction truly named "nop" and not taking any operands.
Many architectures don't have such. But this has been overcome by
the introduction of the .nop directive, which merely requires the
arch to have some insn which is "no operation" (which might e.g.
be ORin zero into a register). I'm pretty sure almost all
architectures will have such. I'm also quite sure though that there
are old tests which are yet to be converted ...

You move but otherwise retain the sentence regarding the need to
change expectations when objdump output style changes. I think
this should be relaxed at least as slightly as saying "may need to
be" instead of "must". Generally you'll find that more modern tests
write expectations in suitably relaxed regexp-s such that quite a
few possible layout/style changes wouldn't require adjustments. And
I think if doc is adjusted in this area anyway, it should recommend
to write expectations as tight as necessary (for the purpose of
the test), but as relaxed as possible.

Jan