Re: [RFA PATCH 2/3] Add debug-stabs debug-dwarf and class option for pascal compiler

[Doug Evans <dje at google dot com>]

On Fri, Jan 9, 2015 at 4:55 PM, Pierre Muller
<pierre.muller@ics-cnrs.unistra.fr> wrote:
>> Hi.
>> This patch makes me uncomfortable.
>> It's to a pascal specific file, so at least the discomfort is contained
>> :-),
>> but I wouldn't want this spreading.
>>
>> How to select debug information should be
>> orthogonal to compilation language,
>> and there are a myriad of ways to select what kind of dwarf debug info
>> to get (with/without type units, with/without .gdb_index, with/without
>> dwz, and so on).  I have board files to help me drive the various
>> combinations I'm interested in.  This won't work if gdb.pascal/*
>> starts hardcoding debug info into the test.
>>
>> Maybe there's a good reason to do it this way for pascal,
>> but I need more data.
>
>  Hi Doug, the reasons of this patch are:
>
> 1) the particular problem that the series fixes is strictly limited
> to stabs debugging information.
>
>   Thus to really test that this bug has been fixed, I need to compile
> the test code using stabs debugging format (which I now realize
> I did not do in the third part of the patch series).
>
> 2) the pascal language support tries to support both
> GNU pascal compiler (aka GPC) and Free Pascal Compiler (aka FPC).
>
>   These two compilers have very different options,
> so the submitted patch purpose is to unify the selection of stabs versus dwarf
> debug format independently from the used pascal compiler.
>
>  For other pascal tests, only debug is used, which results in the default
> format according to the target architecture and operating system.
>
> I hope that the explanations above are sufficient to ease your discomfort.

Yeah, I figured those are the reasons, but I dunno.
We're inventing something new here when existing mechanisms
can already handle this.

One *could* have the test be debug-format agnostic,
and to test with stabs just do
make check RUNTESTFLAGS=--target_board=stabs

If you're using stabs you'll want to run the whole testsuite with stabs anyway.

Plus if a compiler has a different spelling for -gstabs, one could
have another board file for that compiler: you'll want to run the
whole testsuite
with that compiler anyway.

make check RUNTESTFLAGS="gdb.pascal/*.exp --target_board=fpc-stabs"

or some such.

Obviously, by "whole testsuite" I mean the pascal parts.
So that raises a question that may help guide the discussion.
When testing with fpc, how do you run the testsuite?
Do you just run the pascal subset, gdb.pascal/*.exp?
Or do you run the WHOLE testsuite (i.e., gdb.*/*.exp),
e.g., just a plain "make check"
and expect the test harness to pick up fpc and DTRT.
How do other languages handle this I wonder.

I'm not totally opposed to this (I don't like to say "No."),
but I'm still uncomfortable with the patch.