[Patch v6 4/6] restore: transparently support gzipped input

[Austin Clements]

Quoth David Bremner on Apr 03 at  4:41 pm:
> We rely completely on zlib to do the right thing in detecting gzipped
> input. Since our dump format is chosen to be 7 bit ascii, this should
> be fine.
> ---
>  doc/man1/notmuch-restore.rst |  8 ++++++++
>  notmuch-restore.c            | 41 ++++++++++++++++++++++++++---------------
>  test/T240-dump-restore.sh    | 14 ++++++++++++++
>  3 files changed, 48 insertions(+), 15 deletions(-)
> 
> diff --git a/doc/man1/notmuch-restore.rst b/doc/man1/notmuch-restore.rst
> index d6cf19a..936b138 100644
> --- a/doc/man1/notmuch-restore.rst
> +++ b/doc/man1/notmuch-restore.rst
> @@ -50,6 +50,14 @@ Supported options for **restore** include
>              format, this heuristic, based the fact that batch-tag format
>              contains no parentheses, should be accurate.
>  
> +GZIPPED INPUT
> +=============
> +
> +\ **notmuch restore** will detect if the input is compressed in
> +**gzip(1)** format and automatically decompress it while reading. This
> +detection does not depend on file naming and in particular works for
> +standard input.
> +
>  SEE ALSO
>  ========
>  
> diff --git a/notmuch-restore.c b/notmuch-restore.c
> index c54d513..eb5b7b2 100644
> --- a/notmuch-restore.c
> +++ b/notmuch-restore.c
> @@ -22,6 +22,7 @@
>  #include "hex-escape.h"
>  #include "tag-util.h"
>  #include "string-util.h"
> +#include "zlib-extra.h"
>  
>  static regex_t regex;
>  
> @@ -128,10 +129,9 @@ notmuch_restore_command (notmuch_config_t *config, int argc, char *argv[])
>      tag_op_list_t *tag_ops;
>  
>      char *input_file_name = NULL;
> -    FILE *input = stdin;
> +    gzFile input;
>      char *line = NULL;
>      void *line_ctx = NULL;
> -    size_t line_size;
>      ssize_t line_len;
>  
>      int ret = 0;
> @@ -163,13 +163,23 @@ notmuch_restore_command (notmuch_config_t *config, int argc, char *argv[])
>      if (! accumulate)
>  	flags |= TAG_FLAG_REMOVE_ALL;
>  
> -    if (input_file_name) {
> -	input = fopen (input_file_name, "r");
> -	if (input == NULL) {
> -	    fprintf (stderr, "Error opening %s for reading: %s\n",
> -		     input_file_name, strerror (errno));
> +    if (input_file_name)
> +	input = gzopen (input_file_name, "r");
> +    else {
> +	int infd = dup (STDIN_FILENO);
> +	if (infd < 0) {
> +	    fprintf (stderr, "Error duping stdin\n");
>  	    return EXIT_FAILURE;
>  	}
> +	input = gzdopen (infd, "r");
> +	if (! input)
> +	    close (infd);
> +    }
> +
> +    if (input == NULL) {
> +	fprintf (stderr, "Error opening %s for (gzip) reading: %s\n",
> +		 input_file_name ? input_file_name : "stdin", strerror (errno));

There's a sketchy line about errno in the gz(d)open docs: "On error,
gzopen() *may* set the global variable errno to indicate the error."
(emphasis mine).  This suggests we should set errno to 0 before the
calls to gz(d)open above.

> +	return EXIT_FAILURE;
>      }
>  
>      if (opt_index < argc) {
> @@ -184,12 +194,17 @@ notmuch_restore_command (notmuch_config_t *config, int argc, char *argv[])
>      }
>  
>      do {
> -	line_len = getline (&line, &line_size, input);
> +	util_status_t status;
> +
> +	status = gz_getline (line_ctx, &line, &line_len, input);
>  
>  	/* empty input file not considered an error */
> -	if (line_len < 0)
> +	if (status == UTIL_EOF)
>  	    return EXIT_SUCCESS;
>  
> +	if (status)

Will this lead to a silent exit failure if there's a problem with
decompression?  This suggests we should have a UTIL_GZERROR that tells
the caller to consult gzerror for the error message.  (Though this is
still an improvement over the original code, which would silently
succeed when getline failed!)

> +	    return EXIT_FAILURE;
> +
>      } while ((line_len == 0) ||
>  	     (line[0] == '#') ||
>  	     /* the cast is safe because we checked about for line_len < 0 */
> @@ -254,7 +269,7 @@ notmuch_restore_command (notmuch_config_t *config, int argc, char *argv[])
>  	if (ret)
>  	    break;
>  
> -    }  while ((line_len = getline (&line, &line_size, input)) != -1);
> +    }  while (gz_getline (line_ctx, &line, &line_len, input) == UTIL_SUCCESS);

It looks like a gz_getline error here will cause restore to stop and
claim that the restore was successful.  (The original code has this
problem, too.)

>  
>      if (line_ctx != NULL)
>  	talloc_free (line_ctx);
> @@ -262,13 +277,9 @@ notmuch_restore_command (notmuch_config_t *config, int argc, char *argv[])
>      if (input_format == DUMP_FORMAT_SUP)
>  	regfree (&regex);
>  
> -    if (line)
> -	free (line);
> -
>      notmuch_database_destroy (notmuch);
>  
> -    if (input != stdin)
> -	fclose (input);
> +    gzclose_r (input);
>  
>      return ret ? EXIT_FAILURE : EXIT_SUCCESS;
>  }
> diff --git a/test/T240-dump-restore.sh b/test/T240-dump-restore.sh
> index b6d8602..efe463e 100755
> --- a/test/T240-dump-restore.sh
> +++ b/test/T240-dump-restore.sh
> @@ -80,6 +80,20 @@ notmuch dump --gzip --output=dump-gzip-outfile.gz
>  gunzip dump-gzip-outfile.gz
>  test_expect_equal_file dump.expected dump-gzip-outfile
>  
> +test_begin_subtest "restoring gzipped stdin"
> +notmuch dump --gzip --output=backup.gz
> +notmuch tag +new_tag '*'
> +notmuch restore < backup.gz
> +notmuch dump --output=dump.actual
> +test_expect_equal_file dump.expected dump.actual
> +
> +test_begin_subtest "restoring gzipped file"
> +notmuch dump --gzip --output=backup.gz
> +notmuch tag +new_tag '*'
> +notmuch restore --input=backup.gz
> +notmuch dump --output=dump.actual
> +test_expect_equal_file dump.expected dump.actual
> +
>  # Note, we assume all messages from cworth have a message-id
>  # containing cworth.org
>