From: René Scharfe Date: Tue, 16 Jun 2026 22:14:37 GMT Subject: Re: [PATCH] cat-file: speed up default format Message-ID: <47e3cf16-217e-45d4-91e2-5a1abb4ee49e@web.de> In-Reply-To: <20260616111237.GA687438@coredump.intra.peff.net> On 6/16/26 1:12 PM, Jeff King wrote: > On Mon, Jun 15, 2026 at 11:53:07PM +0200, René Scharfe wrote: > >> We can also store literal characters in there. An opcode plus with a >> payload char incurs an overhead of 50%, which sounds high, but at least >> the default format only has two of them and it's much better than >> storing pointer plus size for an overhead of more than 90% in case of a >> single char. > > True, and it's a size win if the literal portions tend to be small > (fewer than 15 bytes). You do lose out on the ability to strbuf_add() > them in one go, though. So lots more strbuf_grow() checks, etc. If you > really wanted to get fancy, you could follow the opcode with a length > represented as a variable-sized integer, followed by the literal bytes. Or an opcode that shovels a fixed-size string. Depends on how much literal text people include in their formats. > I'm not sure that Git's formatting code needs to squeeze out quite that > much performance, though. Good point. Near-native performance would be necessary to make peephole optimizations like the special handling of the default format unnecessary, which I understand exists to speed up Gitaly [1], but I guess most users don't have such high demands. And there's no point in removing a few lines of duplicate code if the necessary machinery adds a lot of complexity. Though the code discussed so far was not too crazy IMHO. [1] https://gitlab.com/gitlab-org/gitaly/-/blob/master/internal/git/gitpipe/catfile_info.go > OK, so we managed another 1%. But I'm skeptical that this linear opcode > technique is where we want to go in the long run, if we're ever going to > unify formatters. Agreed. René