[COMMITTED 2/3] gas: BPF pseudo-c syntax tests

[Jose E. Marchesi]

> On 26.04.2023 19:31, Jose E. Marchesi via Binutils wrote:
>> --- a/gas/ChangeLog
>> +++ b/gas/ChangeLog
>> @@ -1,3 +1,23 @@
>> +2023-04-20  Guillermo E. Martinez  <guillermo.e.martinez@oracle.com>
>> +
>> +	PR gas/29728
>> +	* testsuite/gas/all/assign-bad-recursive.d: Skip test in bpf-*
>> +	targets.
>> +	* testsuite/gas/all/eqv-dot.d: Likewise.
>> +	* testsuite/gas/all/gas.exp: Skip other assignment tests in bpf-*.
>
> I view doing such as problematic. Looking at what patch 3 documents,
> the uses of " = " are pretty limited, and ones not naming a register
> on the lhs (or, for store forms, on the rhs) ought to be fine to
> retain their meaning. Sadly there isn't an easy way to specify target-
> specific flags, or else I'd be inclined to suggest that you have an
> option to suppress recognition of the C-like syntax (which may be a
> good idea anyway, as people might be using constructs like the ones
> used in the testcases you now disable) and use it here and below.

Hmm, actually the assembler is supposed to first try to recognize normal
syntax, including expressions such as SYMBOL = VALUE.  Then to fallback
to the pseudo-c syntax in case the above fails to parse.

I will take a look (Guillermo is no longer working with us.)  It may be
that we don't need to disable these tests at all.

> In any event, ...
>
>> --- a/gas/testsuite/gas/all/eqv-dot.d
>> +++ b/gas/testsuite/gas/all/eqv-dot.d
>> @@ -2,7 +2,7 @@
>>  #name: eqv involving dot
>>  # bfin doesn't support 'symbol = expression'
>>  # tic30 and tic4x have 4 octets per byte, tic54x has 2 octets per byte
>> -#notarget: bfin-*-* *c30-*-* *c4x-*-* *c54x-*-*
>> +#notarget: bfin-*-* *c30-*-* *c4x-*-* *c54x-*-* *bpf-*-*
>
> ... at least in cases where there already are justifying comments, new
> additions of exceptions shouldn't go uncommented.

Yes, will do, in case it is necessary to keep the skips.

>> --- a/gas/testsuite/gas/bpf/alu-be.d
>> +++ b/gas/testsuite/gas/bpf/alu-be.d
>> @@ -1,5 +1,6 @@
>>  #as: --EB
>>  #source: alu.s
>> +#source: alu-pseudoc.s
>>  #objdump: -dr
>>  #name: eBPF ALU64 instructions, big endian
>
> I may of course be reading binutils-common.exp's run_dump_test wrong,
> but is this having the intended effect of assembling each of the files
> once and checking objdump output for each of them? It looks to me as
> if only the assembling step would be performed for both, which I don't
> think is what is wanted.

It was an attempt to avoid having to replicate the same contents in
alu-be.d and alu-be-pseudoc.d.  Will look into this too.

>> --- a/gas/testsuite/gas/macros/macros.exp
>> +++ b/gas/testsuite/gas/macros/macros.exp
>> @@ -82,6 +82,7 @@ switch -glob $target_triplet {
>>      rl78-*-* { }
>>      rx-*-* { }
>>      vax-*-* { }
>> +    bpf-*-* { }
>>      default { run_list_test dot "-alm" }
>>  }
>
> How is this test affected by your changes? It consists of only labels
> and directives afaics, so insn syntax expectations shouldn't matter
> at all.
>
> Jan