Re: [PATCH 1/4] strbuf: fix doc for `strbuf_attach()`

From: Junio C Hamano <gitster@pobox.com>
To: "Martin Ågren" <martin.agren@gmail.com>
Cc: git@vger.kernel.org, "Đoàn Trần Công Danh" <congdanhqx@gmail.com>
Subject: Re: [PATCH 1/4] strbuf: fix doc for `strbuf_attach()`
Date: Mon, 20 Apr 2020 10:30:31 -0700	[thread overview]
Message-ID: <xmqqd082rrns.fsf@gitster.c.googlers.com> (raw)
In-Reply-To: <54e10bb06a39d23591ea4d02665083d705682468.1587297254.git.martin.agren@gmail.com> ("Martin =?utf-8?Q?=C3=85gren=22's?= message of "Sun, 19 Apr 2020 14:32:27 +0200")

Martin Ågren <martin.agren@gmail.com> writes:

> Commit 917c9a7133 ("New strbuf APIs: splice and attach.", 2007-09-15)
> added `strbuf_attach()`. In the commit message, it is explicitly
> discussed how using `mem == len` is ok, although possibly costly. When
> this function was documented in dd613e6b87 ("Strbuf documentation:
> document most functions", 2008-06-04), this distinction was missed and
> the documentation simply forbids this case.

In other words, mem==len implies that the original lacks the
terminating '\0', so attach needs to reallocate from the get go, so
discouraging such use may make sense, but it is too strong to forbid
it, as the strbuf invariant is held safely?  If so, the description
makes sense to me.

> The function handles reallocation and truncation, yet the docs say that
> the "amount [of allocated memory] must be larger than the string length,
> because the string you pass is supposed to be a NUL-terminated string".

IOW, _attach() does not mind if the original lacks '\0' at the end.

> diff --git a/strbuf.h b/strbuf.h
> index ce8e49c0b2..2a462f70cc 100644
> --- a/strbuf.h
> +++ b/strbuf.h
> @@ -112,10 +112,12 @@ char *strbuf_detach(struct strbuf *sb, size_t *sz);
>  /**
>   * Attach a string to a buffer. You should specify the string to attach,
>   * the current length of the string and the amount of allocated memory.
> + * The amount must be at least as large as the string length. If the two
> + * lengths are equal, reallocation will be handled as appropriate and in
> + * any case, the string will be NUL-truncated as implied by `len`.

NUL-truncated?  Ah, if mem and len are the same, the string is reallocated
to fit an extra byte to NUL-terminate, to make sure strlen(sb->buf)==len
holds.  Makes sense.

> + *
> + * This string _must_ be malloc()ed, and after attaching, the pointer
> + * cannot be relied upon anymore, and neither be free()d directly.

That's a good thing to explain.

>   */
>  void strbuf_attach(struct strbuf *sb, void *str, size_t len, size_t mem);