Paolo Bonzini <[email protected]> writes:
> On 25/11/2015 15:47, Markus Armbruster wrote:
>> > This patch from 2011 (!) saves about 96% of the allocation cost (down
>> > from 500 MiB to 20 MiB) for check-qjson.
>>
>> Looks good to me apart from the incomplete transition to recursive
>> descent in PATCH 2.
>
> I do not understand whether that is a problem. Is the remark from the
> commit message incorrect ("Saving the parser context is mostly
> unnecessary; we can replace it with peeking at the next token, or remove
> it altogether when the restore only happens on errors. The token list
> is destroyed anyway on errors.")?
The token sequence
{ 1 [ 2 ] }
gets parsed as [ 2 ]. See my reply to PATCH 2 for a more detailed
explanation.
I didn't bother to construct an example where QMP executes the incorrect
result, and I'm not sure it's possible.
Regardless, I'm fixing it, by simplifying the parser some more.
