Re: [PATCH] http-backend: Treat empty CONTENT_LENGTH as zero

From: Jonathan Nieder <jrnieder@gmail.com>
Date: Sep 11, 2018, 02:20 UTC
Message-ID: <20180911022028.GA20518@aiede.svl.corp.google.com>
In-Reply-To: <20180911015553.GA5838@sigill.intra.peff.net>

Jeff King wrote:

> On Mon, Sep 10, 2018 at 02:22:21PM -0700, Jonathan Nieder wrote:

Show 12 quoted lines

>> Thanks.  I am wondering if we should go all the way and do
>>
>> 	ssize_t val;
>> 	const char *str = getenv("CONTENT_LENGTH");
>>
>> 	if (!str || !*str)
>> 		return 0;
>> 	if (!git_parse_ssize_t(str, &val))
>> 		die(...);
>> 	return val;
>>
>> That would match the RFC, but it seems to make t5510-fetch.sh hang,

[...]

Show 13 quoted lines

>> Do you know why?
>
> Yes. :)
>
> It's due to this comment in the patch you are replying to:
>
> +       if (!str) {
> +               /*
> +                * RFC3875 says this must mean "no body", but in practice we
> +                * receive chunked encodings with no CONTENT_LENGTH. Tell the
> +                * caller to read until EOF.
> +                */
> +               val = -1;

Ah! So "in practice" includes "in Apache". An old discussion[1] on Apache's httpd-users list agrees.

The question then becomes: what does IIS do for zero-length requests? Does any other web server fail to support "read until EOF" in general?

The CGI standard does not cover chunked encoding so we can't lean on the standard for advice. It's not clear to me yet whether this patch improves on what's in "master".

Thanks, Jonathan

[1] http://mail-archives.apache.org/mod_mbox/httpd-users/200909.mbox/%3C4AAACC38.3070200@rowe-clan.net%3E