[PATCH 5 of 7 V3] sparse-revlog: introduce native (C) implementation of slicechunktodensity

[Boris FELD]

On 19/11/2018 13:54, Yuya Nishihara wrote:
> On Mon, 19 Nov 2018 10:42:17 +0100, Boris Feld wrote:
>> # HG changeset patch
>> # User Boris Feld <boris.feld at octobus.net>
>> # Date 1542276598 -3600
>> #      Thu Nov 15 11:09:58 2018 +0100
>> # Node ID b365764b9915e600a5c7c273f244f65e2240e00a
>> # Parent  f9ce23d5d3aeec9a9e8589ec83d089b2df83da82
>> # EXP-Topic sparse-perf
>> # Available At https://bitbucket.org/octobus/mercurial-devel/
>> #              hg pull https://bitbucket.org/octobus/mercurial-devel/ -r b365764b9915
>> sparse-revlog: introduce native (C) implementation of slicechunktodensity
>> +		Py_ssize_t revstart;
>> +		Py_ssize_t revsize;
>> +		if (!index_get_start(self, revs[i], &revstart)) {
>> +			goto bail;
> Be aware that Py_ssize_t * != long *. I think Py_ssize_t is the right type
> here and other places computing offsets or indices.
>
> There are a few more callers which passes Py_ssize_t * as long *argument.

Looking more closely to this, it seems like offset can be full unsigned
64 bits integer. Py_ssize_t can be smaller than that. So I think we
should keep offset, length and segmentspan as explicit uint64_t. And use
PyLong_AsUnsignedLongLong if conversion is needed.

Does it seem right to you? shall I update the series in that directory?

Because we use unsigned type, it is inconvenient to return -1 value.
That is why I went for the boolean (same as pylong_to_long).

Can you confirm the following:
1) using uint64 for data offset and length
2) using Py_ssize_t for everything else
3) Handle error as booleans (using true/false)

Cheers,

> _______________________________________________
> Mercurial-devel mailing list
> Mercurial-devel at mercurial-scm.org
> https://www.mercurial-scm.org/mailman/listinfo/mercurial-devel