How to pass a filename argument gitconfig diff textconv?

Question

The documentation on textconv at https://git.wiki.kernel.org/index.php/Textconv has the succinct wording:

~/.gitconfig must indicate the command to execute for the the textconv driver:

[diff "<driver_name>"]
    textconv=<command>

I cannot find any documentation on how to format a command which requires the filename passed as a positional argument. For example, I'd like to use the following pdf formatter command, which requires a dash as last argument to write to stdout:

[diff "pdf"]
    textconv = pdftotext -layout "$1" -

For now I've had to write custom one-liner shell-scripts as a workaround, but they start to accumulate, and it becomes a bit annoying.

Is there a way to do without those scripts? The "$1" or xargs' '{}' convention for arguments don't seem to work.

Answer 1

As a workaround for gitconfig lacking parameter substitution functionality, you can wrap the command into a shell call, as in

[diff "pdf"]
    textconv = sh -c 'pdftotext -layout -enc UTF-8 "$0" -'

(found here: https://gist.github.com/t-yuki/9348e5d4aa4a75a6acf9)

Answer 2

Unfortunately, that information is just plain not available. Here is the code that actually achieves the text conversion:

temp = prepare_temp_file(r, spec->path, spec);
*arg++ = pgm;
*arg++ = temp->name;
*arg = NULL;


child.use_shell = 1;
child.argv = argv;
child.out = -1;
if (start_command(&child)) {
    remove_tempfile();
    return NULL;
}

The two arguments are the name of the program itself (argv[0] as usual) and the name of the temporary file containing the bits extracted from wherever they reside (based on the spec parameter; follow the link above for additional detail).

The spec probably does carry the original path name in most or all cases (and if not, look at the function just below this one), it's just not copied through to the program arguments. The textconv filter code could, but doesn't, employ the %-expansion technique used by Git merge drivers, and if it did (but it doesn't) it could has a %-escape that passed the original file name. But of course it doesn't.

(You can make your own clone of Git and work on it, and perhaps try to convince the Git folks to take your change as a contribution to public Git...)