Re: [PATCH] searchidx: index "diff --git a/... b/..." headers

[Rob Herring <robh@kernel.org>]

On Mon, Nov 8, 2021 at 9:12 PM Eric Wong <e@80x24.org> wrote:
>
> Rob Herring <robh@kernel.org> wrote:
> > On Mon, Nov 8, 2021 at 3:27 PM Eric Wong <e@80x24.org> wrote:
> > >
> > > Rob Herring <robh@kernel.org> wrote:
> > > > On Mon, Nov 8, 2021 at 2:22 PM Konstantin Ryabitsev
> > > > > I think 's:patch AND nq:diff' is a good option here.
> > > >
> > > > Not even close really. That mainly finds my replies with 'diff' in
> > > > them. I'm not sure why, but it misses most actual patches:
> > > >
> > > > https://lore.kernel.org/all/?q=s%3Apatch+nq%3Adiff+f%3Arobh%40kernel.org
> > >
> > > Actually, it looks like nq:diff never works.  The diff indexer
> > > skips right over 'diff --git a/... b/...' lines :x
> >
> > Never works for 'diff' being a patch? Because it works very well
> > finding all the other cases.
>
> Yeah, the index_diff() code path ignored the "diff --git" phrase
> before this patch.
>
> > > The following should fix it, but reindexing is necessary.
> > > ---------8<----------
> > > Subject: [PATCH] searchidx: index "diff --git a/... b/..." headers
> > >
> > > While we do detailed indexing of git diffs, the header itself
> > > was failing and queries like 'nq:diff' would not work.
> >
> > Any thoughts on supporting an 'is a patch' type query?
>
> I think 's:patch' should be sufficient, don't think there's
> many false-positives on that front, actually.

It's at least 's:patch OR s:rfc OR s:resend'. That catches all but the
few creative folks that come up with something else.

> With this fix, nq:"diff --git" should also be working across
> https://yhbt.net/lore/ in about 40 hours (whenever reindex
> finishes)

'diff --git' should cover probably 99.9% of patches but there are
still some non-git diffs from time to time.

> I'm not sure if there needs to be a specific term to index
> patches on; maybe there is.  There's still a lot of Xapian
> we're not using, yet...

What I'm hoping to get to is a replacement for patchwork in my
workflow. For that I want all patches which don't have either a
Reviewed/Acked tag from me or a reply from me. I think the first part
should be possible with lei, but I'd imagine the last part is some
processing on top of the lei query.

Rob