> I haven't looked at the performance side of this.  It just seemed simpler.

I say we keep it as-is, as my reading of the current commands is simpler
than the way I read this alternative.  Compare the way I read it: "Make
a tar archive and list its contents."  Versus: "Execute tar with verbose
output, which happens to only include the files packaged, also that
output needs to be redirected from stderr to stdout for make to process
it, also bury the archive contents in /dev/null so as not to interfere
with the redirected verbose output."

> No, I was rather wondering if this behavior is documented or just accidental.

Although I do not see an example of an explicitly mentioned file
argument used with an exclusion flag in the tar manual, the language of
the manual suggests that this should work:

>  `--exclude-from=file'
> `-X file'
> 
>     Causes tar to ignore files that match the patterns listed in file. 

"Files" ought to include explicitly mentioned files.

Also, I've attached a patch for the "pwd-based exclusions" issue I
mentioned in my last email.  I checked the timing on it, and it only
reduces the speed of "make" with nothing to do by about 50ms.  Does it
look good to you?

Jackson

On 12/25/2016 08:16 PM, Stefan Monnier wrote:
>>> I think we can skip the second tar with something like
>>> "tar -cvhf /dev/null 2>&1".
>> I tried this before, and again just now, but it does not cause make to
>> execute at a significantly different speed on my machine.  Is the effect
>> more noticeable on yours?
> 
> I haven't looked at the performance side of this.  It just seemed simpler.
> 
>> $ cd a
>> $ tar -ch *.el --no-recursion --exclude-vcs -X .elpaignore | tar --list
>>
>> The above command correctly yields "b.el".
>>
>> Did you have any other cases in mind?
> 
> No, I was rather wondering if this behavior is documented or just accidental.
> 
> 
>         Stefan
>