Re: JSON/YAML/TOML/etc. parsing performance

[Philipp Stephani <p.stephani2@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1572 bytes --]

Philipp Stephani <p.stephani2@gmail.com> schrieb am So., 29. Okt. 2017 um
23:49 Uhr:

> Philipp Stephani <p.stephani2@gmail.com> schrieb am So., 29. Okt. 2017 um
> 21:48 Uhr:
>
>> Paul Eggert <eggert@cs.ucla.edu> schrieb am Mo., 9. Okt. 2017 um
>> 08:19 Uhr:
>>
>>> Philipp Stephani wrote:
>>> > I don't think Jansson can use xmalloc because xmalloc can exit
>>> nonlocally,
>>> > which is not expected by a third-party library such as Jansson. It
>>> could
>>> > use a suitable wrapper of lmalloc, though.
>>>
>>> That would be overkill, as lmalloc arranges for Lisp alignment, which
>>> Jansson
>>> does not need. We could define new functions (smalloc and srealloc,
>>> say), that
>>> act like malloc and realloc except they return NULL for requests larger
>>> than
>>> PTRDIFF_MAX. Right now, I expect only the JSON code needs this sort of
>>> thing so
>>> we could put the new functions in json.c. If other code needs it later
>>> we could
>>> move these new functions to alloc.c.
>>>
>>
>> Yes, that sounds reasonable.
>>
>
> Here's a new patch that incorporates some of these changes. Specifically:
>
> - I've removed some of the assertions
> - I've installed a custom allocator, as you suggested
> - Reverted back to creating a temporary string and inserting that into the
> buffer. Anything else just doesn't seem to work or seems way too complex.
> - Introduced explicit encoding and decoding. I suspect that will lead to a
> massive performance hit, but I haven't done any benchmarks yet.
> - Added manual section and NEWS entry
>

Is that patch OK for master?

[-- Attachment #2: Type: text/html, Size: 2547 bytes --]