git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH 4/5] archive-tar: stream large blobs to tar file

From
Junio C Hamano <gitster@pobox.com>
Date
Apr 30, 2012, 21:36 UTC
Message-ID
<7vhaw0ewb8.fsf@alter.siamese.dyndns.org>
In-Reply-To
<4F9EFF39.4010804@lsrfire.ath.cx>
René Scharfe <rene.scharfe@lsrfire.ath.cx> writes:
Show 54 quoted lines
>> +static int stream_blob_to_file(const unsigned char *sha1)
>> +{
>> +	struct git_istream *st;
>> +	enum object_type type;
>> +	unsigned long sz;
>> +
>> +	st = open_istream(sha1,&type,&sz, NULL);
>> +	if (!st)
>> +		return error("cannot stream blob %s", sha1_to_hex(sha1));
>> +	for (;;) {
>> +		char buf[BLOCKSIZE];
>> +		ssize_t readlen;
>> +
>> +		readlen = read_istream(st, buf, sizeof(buf));
>> +
>> +		if (readlen <= 0)
>> +			return readlen;
>> +		write_blocked(buf, readlen);
>> +	}
>> +	close_istream(st);
>> +	return 0;
>> +}
>
> The stream is never closed.  Perhaps squash this in?
>
> diff --git a/archive-tar.c b/archive-tar.c
> index 506c8cb..6109fd3 100644
> --- a/archive-tar.c
> +++ b/archive-tar.c
> @@ -66,6 +66,7 @@ static void write_blocked(const void *data, unsigned long size)
>  static int stream_blob_to_file(const unsigned char *sha1)
>  {
>  	struct git_istream *st;
> +	ssize_t readlen;
>  	enum object_type type;
>  	unsigned long sz;
>  
> @@ -74,16 +75,15 @@ static int stream_blob_to_file(const unsigned char *sha1)
>  		return error("cannot stream blob %s", sha1_to_hex(sha1));
>  	for (;;) {
>  		char buf[BLOCKSIZE];
> -		ssize_t readlen;
>  
>  		readlen = read_istream(st, buf, sizeof(buf));
>  
>  		if (readlen <= 0)
> -			return readlen;
> +			break;
>  		write_blocked(buf, readlen);
>  	}
>  	close_istream(st);
> -	return 0;
> +	return readlen;
>  }

Your patch on top obviouly is the right thing to do, but reading the code again, I am not sure if the original is correct. read_istream() itself does not promise that it will always fill the buffer before returning (it could return with a short read). It seems incorrect that the caller does not loop to avoid padding a short read with padding by calling write_blocked().

Previous: René ScharfeNext: René Scharfe
Message 11 of 23 in “Large file support for git-archive”
  1. 0/5 Large file support for git-archiveNguyễn Thái Ngọc Duy, Apr 30, 2012
  2. 1/5 archive-tar: turn write_tar_entry into blob-writing onlyNguyễn Thái Ngọc Duy, Apr 30, 2012
  3. Junio C HamanoApr 30, 2012
  4. René ScharfeApr 30, 2012
  5. 2/5 archive-tar: unindent write_tar_entry by one levelNguyễn Thái Ngọc Duy, Apr 30, 2012
  6. 3/5 archive: delegate blob reading to backendNguyễn Thái Ngọc Duy, Apr 30, 2012
  7. René ScharfeApr 30, 2012
  8. 4/5 archive-tar: stream large blobs to tar fileNguyễn Thái Ngọc Duy, Apr 30, 2012
  9. Junio C HamanoApr 30, 2012
  10. René ScharfeApr 30, 2012
  11. Junio C HamanoApr 30, 2012
  12. René ScharfeApr 30, 2012
  13. 5/5 archive-zip: stream large blobs into zip fileNguyễn Thái Ngọc Duy, Apr 30, 2012
  14. Junio C HamanoApr 30, 2012
  15. René ScharfeApr 30, 2012
  16. streaming: void pointer instead of char pointerRené Scharfe, Apr 30, 2012
  17. archive-zip: remove uncompressed_sizeRené Scharfe, Apr 30, 2012
  18. archive-zip: factor out helpers for writing sizes and CRCRené Scharfe, Apr 30, 2012
  19. archive-zip: streaming for stored filesRené Scharfe, Apr 30, 2012
  20. archive-zip: streaming for deflated filesRené Scharfe, Apr 30, 2012
  21. Junio C HamanoApr 30, 2012
  22. René ScharfeApr 30, 2012
  23. Nguyen Thai Ngoc DuyMay 1, 2012

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.