Re: RFA: Fix readelf -wi test (Was: Re: readelf: apply all rela relocations in the debug_info section)

[Joern Rennecke <joern dot rennecke at superh dot com>]

Nick Clifton wrote:
> Your patch still does not resolve the "readelf: Warning: Skipping
> unexpected symbol type 1" problem that you mentioned in your earlier
> post.  In my opinion however the readelf.wi is severely broken, since
> there is no real guaranteed structure to the output of "readelf -wi".
> 
> Instead I am going to apply your patch together with a change of my
> own.  This additional patch:
> 
>   * Filters out "skipping unexpected symbol" messages.
> 
>   * Shortens the expected output file (readelf.wi) so that only the
>     first few blocks are present.  These blocks ought to be in every
>     toolchain's output.  They were for all the toolchains I tested.
> 
>   * Adds several more toolchains to the expected failure list, since
>     they do not default to generating dwarf2 output.
> 
>   * Updates testprog.c to explicitly make the char array unsigned, and
>     to remove the inclusion of stdio.h and string.h so that it can be
>     compiled without a complete run time environment present.

Yes, the binutils testsuite passes now on sh-elf.


"H. J. Lu" wrote:	
> Now it fails on RedHat 9:

Not to say that this is acceptable, but if it ever suceeded in recent
years, that must have been a bug, since the line numbers in readelf.wi
were out of sync with testprog.c

The current failure really comes down to the brokenness that Nick
mentioned above.  Either we have to process the output more
intelligently - with GNU grep, we could just use grep -A xx -B yy to
look for the expected bits, paste this all together into a massaged
output file, and compare that to our template; an awk wizard can
surely do better.  Or we could start to supply different templates
for different targets.

-- 
--------------------------
SuperH (UK) Ltd.
2410 Aztec West / Almondsbury / BRISTOL / BS32 4QX
T:+44 1454 465658