[PATCH] http.postbuffer: make a size_t

[PATCH] http.postbuffer: make a size_t

[David Turner]

Unfortunately, in order to push some large repos, the http postbuffer
must sometimes exceed two gigabytes.  On a 64-bit system, this is OK:
we just malloc a larger buffer.

Signed-off-by: David Turner <dturner@twosigma.com>
---
 cache.h  |  1 +
 config.c | 17 +++++++++++++++++
 http.c   |  2 +-
 3 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/cache.h b/cache.h
index fbdf7a815a..a8c1b65db0 100644
--- a/cache.h
+++ b/cache.h
@@ -1900,6 +1900,7 @@ extern int git_parse_maybe_bool(const char *);
 extern int git_config_int(const char *, const char *);
 extern int64_t git_config_int64(const char *, const char *);
 extern unsigned long git_config_ulong(const char *, const char *);
+extern size_t git_config_size_t(const char *, const char *);
 extern int git_config_bool_or_int(const char *, const char *, int *);
 extern int git_config_bool(const char *, const char *);
 extern int git_config_maybe_bool(const char *, const char *);
diff --git a/config.c b/config.c
index 1a4d85537b..7b706cf27a 100644
--- a/config.c
+++ b/config.c
@@ -834,6 +834,15 @@ int git_parse_ulong(const char *value, unsigned long *ret)
 	return 1;
 }
 
+static size_t git_parse_size_t(const char *value, unsigned long *ret)
+{
+	size_t tmp;
+	if (!git_parse_signed(value, &tmp, maximum_unsigned_value_of_type(size_t)))
+		return 0;
+	*ret = tmp;
+	return 1;
+}
+
 NORETURN
 static void die_bad_number(const char *name, const char *value)
 {
@@ -892,6 +901,14 @@ unsigned long git_config_ulong(const char *name, const char *value)
 	return ret;
 }
 
+size_t git_config_size_t(const char *name, const char *value)
+{
+	unsigned long ret;
+	if (!git_parse_size_t(value, &ret))
+		die_bad_number(name, value);
+	return ret;
+}
+
 int git_parse_maybe_bool(const char *value)
 {
 	if (!value)
diff --git a/http.c b/http.c
index 96d84bbed3..ab6080835f 100644
--- a/http.c
+++ b/http.c
@@ -331,7 +331,7 @@ static int http_options(const char *var, const char *value, void *cb)
 	}
 
 	if (!strcmp("http.postbuffer", var)) {
-		http_post_buffer = git_config_int(var, value);
+		http_post_buffer = git_config_size_t(var, value);
 		if (http_post_buffer < LARGE_PACKET_MAX)
 			http_post_buffer = LARGE_PACKET_MAX;
 		return 0;
-- 
2.11.GIT

Re: [PATCH] http.postbuffer: make a size_t

[Shawn Pearce]

On Thu, Mar 30, 2017 at 1:29 PM, David Turner <dturner@twosigma.com> wrote:
> Unfortunately, in order to push some large repos, the http postbuffer
> must sometimes exceed two gigabytes.  On a 64-bit system, this is OK:
> we just malloc a larger buffer.

I'm slightly curious what server you are pushing to that needs the
entire thing buffered to compute a Content-Length, rather than using
Transfer-Encoding: chunked. Most Git-over-HTTP should be accepting
Transfer-Encoding: chunked when the stream exceeds postBuffer.

RE: [PATCH] http.postbuffer: make a size_t

[David Turner]

GitLab.  I can't speak to our particular configuration of it -- but if you have a specific question about what the config is, I can ask our gitlab admins.

> -----Original Message-----
> From: Shawn Pearce [mailto:spearce@spearce.org]
> Sent: Thursday, March 30, 2017 4:42 PM
> To: David Turner <David.Turner@twosigma.com>
> Cc: git <git@vger.kernel.org>
> Subject: Re: [PATCH] http.postbuffer: make a size_t
> 
> On Thu, Mar 30, 2017 at 1:29 PM, David Turner <dturner@twosigma.com>
> wrote:
> > Unfortunately, in order to push some large repos, the http postbuffer
> > must sometimes exceed two gigabytes.  On a 64-bit system, this is OK:
> > we just malloc a larger buffer.
> 
> I'm slightly curious what server you are pushing to that needs the entire thing
> buffered to compute a Content-Length, rather than using
> Transfer-Encoding: chunked. Most Git-over-HTTP should be accepting
> Transfer-Encoding: chunked when the stream exceeds postBuffer.

Re: [PATCH] http.postbuffer: make a size_t

[Junio C Hamano]

David Turner <dturner@twosigma.com> writes:

> Unfortunately, in order to push some large repos, the http postbuffer
> must sometimes exceed two gigabytes.  On a 64-bit system, this is OK:
> we just malloc a larger buffer.
>
> Signed-off-by: David Turner <dturner@twosigma.com>
> ---
>  cache.h  |  1 +
>  config.c | 17 +++++++++++++++++
>  http.c   |  2 +-
>  3 files changed, 19 insertions(+), 1 deletion(-)
>
> diff --git a/cache.h b/cache.h
> index fbdf7a815a..a8c1b65db0 100644
> --- a/cache.h
> +++ b/cache.h
> @@ -1900,6 +1900,7 @@ extern int git_parse_maybe_bool(const char *);
>  extern int git_config_int(const char *, const char *);
>  extern int64_t git_config_int64(const char *, const char *);
>  extern unsigned long git_config_ulong(const char *, const char *);
> +extern size_t git_config_size_t(const char *, const char *);
>  extern int git_config_bool_or_int(const char *, const char *, int *);
>  extern int git_config_bool(const char *, const char *);
>  extern int git_config_maybe_bool(const char *, const char *);
> diff --git a/config.c b/config.c
> index 1a4d85537b..7b706cf27a 100644
> --- a/config.c
> +++ b/config.c
> @@ -834,6 +834,15 @@ int git_parse_ulong(const char *value, unsigned long *ret)
>  	return 1;
>  }
>  
> +static size_t git_parse_size_t(const char *value, unsigned long *ret)
> +{
> +	size_t tmp;
> +	if (!git_parse_signed(value, &tmp, maximum_unsigned_value_of_type(size_t)))

I am getting these:

config.c:840:2: error: pointer targets in passing argument 2 of 'git_parse_signed' differ in signedness [-Werror=pointer-sign]
  if (!git_parse_signed(value, &tmp, maximum_unsigned_value_of_type(size_t)))
  ^

config.c:753:12: note: expected 'intmax_t *' but argument is of type 'size_t *'
 static int git_parse_signed(const char *value, intmax_t *ret, intmax_t max)
            ^

Changing "size_t tmp" to "intmax_t tmp" squelches it but the maximum
unsigned value of size_t type would probably overflow "intmax_t max"
which is signed, so...

Re: [PATCH] http.postbuffer: make a size_t

[Torsten Bögershausen]

On 30/03/17 22:29, David Turner wrote:
> Unfortunately, in order to push some large repos, the http postbuffer
> must sometimes exceed two gigabytes.  On a 64-bit system, this is OK:
> we just malloc a larger buffer.
>
> Signed-off-by: David Turner <dturner@twosigma.com>
> ---
>  cache.h  |  1 +
>  config.c | 17 +++++++++++++++++
>  http.c   |  2 +-
>  3 files changed, 19 insertions(+), 1 deletion(-)
>
> diff --git a/cache.h b/cache.h
> index fbdf7a815a..a8c1b65db0 100644
> --- a/cache.h
> +++ b/cache.h
> @@ -1900,6 +1900,7 @@ extern int git_parse_maybe_bool(const char *);
>  extern int git_config_int(const char *, const char *);
>  extern int64_t git_config_int64(const char *, const char *);
>  extern unsigned long git_config_ulong(const char *, const char *);
> +extern size_t git_config_size_t(const char *, const char *);
>  extern int git_config_bool_or_int(const char *, const char *, int *);
>  extern int git_config_bool(const char *, const char *);
>  extern int git_config_maybe_bool(const char *, const char *);
> diff --git a/config.c b/config.c
> index 1a4d85537b..7b706cf27a 100644
> --- a/config.c
> +++ b/config.c
> @@ -834,6 +834,15 @@ int git_parse_ulong(const char *value, unsigned long *ret)
>  	return 1;
>  }
>
> +static size_t git_parse_size_t(const char *value, unsigned long *ret)
> +{
> +	size_t tmp;
> +	if (!git_parse_signed(value, &tmp, maximum_unsigned_value_of_type(size_t)))
> +		return 0;
> +	*ret = tmp;
> +	return 1;
> +}
What is the return value here ?
Isn't it a size_t we want ?
(There was a recent discussion about "unsigned long" vs "size_t", which
  are the same on many systems, but not under Win64)
Would the following work ?

static int git_parse_size_t(const char *value, size_t *ret)
{
	if (!git_parse_signed(value, ret, maximum_unsigned_value_of_type(size_t)))
		return 0;
	return 1;
}

[]
> +size_t git_config_size_t(const char *name, const char *value)
> +{
> +	unsigned long ret;
> +	if (!git_parse_size_t(value, &ret))
> +		die_bad_number(name, value);
> +	return ret;
> +}
> +
Same here:
size_t git_config_size_t(const char *name, const char *value)
{
	size_t ret;
	if (!git_parse_size_t(value, &ret))
		die_bad_number(name, value);
	return ret;
}

RE: [PATCH] http.postbuffer: make a size_t

[David Turner]

> -----Original Message-----
> From: Torsten Bögershausen [mailto:tboegi@web.de]
> Sent: Friday, March 31, 2017 12:22 AM
> To: David Turner <David.Turner@twosigma.com>; git@vger.kernel.org
> Subject: Re: [PATCH] http.postbuffer: make a size_t
> 
> 
> 
> On 30/03/17 22:29, David Turner wrote:
> > Unfortunately, in order to push some large repos, the http postbuffer
> > must sometimes exceed two gigabytes.  On a 64-bit system, this is OK:
> > we just malloc a larger buffer.
> >
> > Signed-off-by: David Turner <dturner@twosigma.com>
> > ---
> >  cache.h  |  1 +
> >  config.c | 17 +++++++++++++++++
> >  http.c   |  2 +-
> >  3 files changed, 19 insertions(+), 1 deletion(-)
> >
> > diff --git a/cache.h b/cache.h
> > index fbdf7a815a..a8c1b65db0 100644
> > --- a/cache.h
> > +++ b/cache.h
> > @@ -1900,6 +1900,7 @@ extern int git_parse_maybe_bool(const char *);
> > extern int git_config_int(const char *, const char *);  extern int64_t
> > git_config_int64(const char *, const char *);  extern unsigned long
> > git_config_ulong(const char *, const char *);
> > +extern size_t git_config_size_t(const char *, const char *);
> >  extern int git_config_bool_or_int(const char *, const char *, int *);
> > extern int git_config_bool(const char *, const char *);  extern int
> > git_config_maybe_bool(const char *, const char *); diff --git
> > a/config.c b/config.c index 1a4d85537b..7b706cf27a 100644
> > --- a/config.c
> > +++ b/config.c
> > @@ -834,6 +834,15 @@ int git_parse_ulong(const char *value, unsigned
> long *ret)
> >  	return 1;
> >  }
> >
> > +static size_t git_parse_size_t(const char *value, unsigned long *ret)
> > +{
> > +	size_t tmp;
> > +	if (!git_parse_signed(value, &tmp,
> maximum_unsigned_value_of_type(size_t)))
> > +		return 0;
> > +	*ret = tmp;
> > +	return 1;
> > +}
> What is the return value here ?
> Isn't it a size_t we want ?

Yeah, that should return an int, since it's just "parsed" vs "unparsed".

> (There was a recent discussion about "unsigned long" vs "size_t", which
>   are the same on many systems, but not under Win64) Would the following
> work ?
> 
> static int git_parse_size_t(const char *value, size_t *ret) {
> 	if (!git_parse_signed(value, ret,
> maximum_unsigned_value_of_type(size_t)))
> 		return 0;
> 	return 1;
> }
> 
> []
> > +size_t git_config_size_t(const char *name, const char *value) {
> > +	unsigned long ret;
> > +	if (!git_parse_size_t(value, &ret))
> > +		die_bad_number(name, value);
> > +	return ret;
> > +}
> > +
> Same here:
> size_t git_config_size_t(const char *name, const char *value) {
> 	size_t ret;
> 	if (!git_parse_size_t(value, &ret))
> 		die_bad_number(name, value);
> 	return ret;
> }

In v2 I decided to make this a ssize_t, since it was previously a signed int.  I know we're not using negative values right now, but since nobody has 2**63 bytes of ram anyway, it's probably safe to keep it signed to keep our options open.