Re: [PATCH 2/2] fast-import: duplicate into history rather than passing ownership

From: Jeff King <peff@peff.net>
Date: Aug 26, 2019, 18:42 UTC
Message-ID: <20190826184252.GC23399@sigill.intra.peff.net>
In-Reply-To: <248c5f9f-ba44-6dec-6f30-f7d193bc22bb@web.de>

On Sun, Aug 25, 2019 at 04:21:54PM +0200, René Scharfe wrote:

Show 8 quoted lines

> > You could xstrndup(command_buf.buf, command_buf.len), which would avoid
> > a hidden strlen.
> 
> xstrndup() also searches for NUL, albeit with memchr(3).  xmemdupz()
> would copy without checking.
> 
> I suspect the simplicity of xstrdup() outweighs the benefits of the
> alternatives, but didn't do any measurements..

Yep. I actually started to write xmemdupz() originally then decided it was unnecessarily verbose and a premature optimization.

I wondered after this exchange whether something like:

  char *strbuf_dup(const struct strbuf *sb)
  {
	return xmemdupz(sb->buf, sb->len);
  }

would be a useful general helper. Grepping around it doesn't seem like there are a lot of candidates.

If we really wanted to micro-optimize, we could have cmd_hist store strbufs, and then we could reuse the same buffers over and over without re-allocating. And use strbuf_addbuf(&cmd_hist.buf, &command_buf). :)

-Peff