Re: [PATCH 5/8] decodetree: Allow group covering the entire insn space

[Richard Henderson]

On 6/2/20 7:35 AM, Peter Maydell wrote:
> On Mon, 18 May 2020 at 17:41, Richard Henderson
> <richard.henderson@linaro.org> wrote:
>>
>> This is an edge case for sure, but the logic that disallowed
>> this case was faulty.  Further, a few fixes scattered about
>> can allow this to work.
>>
>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>> ---
>>  ...est1.decode => succ_pattern_group_nest2.decode} |  2 +-
>>  scripts/decodetree.py                              | 14 +++++++++++---
>>  2 files changed, 12 insertions(+), 4 deletions(-)
>>  rename tests/decode/{err_pattern_group_nest1.decode => 
>> succ_pattern_group_nest2.decode} (85%)
> 
>> @@ -978,6 +980,12 @@ def build_tree(pats, outerbits, outermask):
>>          innermask &= i.fixedmask
>>
>>      if innermask == 0:
>> +        # Edge condition: One pattern covers the entire insnmask
>> +        if len(pats) == 1:
>> +            t = Tree(outermask, innermask)
>> +            t.subs.append((0, pats[0]))
>> +            return t
>> +
>>          text = 'overlapping patterns:'
>>          for p in pats:
>>              text += '\n' + p.file + ':' + str(p.lineno) + ': ' + str(p)
> 
> I don't really understand this code, but does the similar
> looking build_size_tree() also need a change to handle a
> length-one pats ?

I don't think so, because in that case we'd exit earlier with

    if onewidth:
        return SizeLeaf(innermask, minwidth)


r~