-p0 vs. -p1 and associated error messages

Thank you for the pointer to

re: Phasing out -p0 patches: http://groups.drupal.org/node/140204

The weird thing is that my git apply will apply both types of patches just fine if I don't specify the -p switch. Why can't we be as tolerant as Git is? Or is my Git smarter than others?

As the OP in #1345240: Patch not to break xhprof reporting when site name contains spaces mentioned, Eclipse will generate -p0 patches, and last time I looked, there was no way to tell it to do -p1. The thread on g.d.o didn't give any rationale for why we needed to standardize on only one of -p0 or -p1:

jbrown:
What's the rational for standardizing on -p1 instead of -p0?
I really hate seeing all those 'a's and 'b's all the time.

rfay:
Git uses -p1 by default
Since we're now using git, it seemed reasonable for people to be able to use "git apply" to apply patches and "git diff" or "git format-patch" to create them.

Yes, that's what the documentation says, but Git accepts both if you omit -p.

I'm still unable to explain to anyone why we're insisting on -p1 instead of leaving it to Git what it wants to accept.

This is crazy! You are right:

HS@NB2 /d/PDev/drupal/d7 (7.x)
$ git apply /d/temp/drupal.demo_p1_patch.patch

HS@NB2 /d/PDev/drupal/d7 (7.x)
$ git apply -R /d/temp/drupal.demo_p1_patch.patch

HS@NB2 /d/PDev/drupal/d7 (7.x)
$ git apply /d/temp/drupal.demo_p0_patch.patch
fatal: git diff header lacks filename information when removing 1 leading pathname components (line 5)

But now look at the following, using http://drupal.org/files/devel_0.patch from #1345240: Patch not to break xhprof reporting when site name contains spaces:

HS@NB2 /d/PDev/drupal/d6/sites/all/modules/devel (6.x-1.x)
$ git apply /d/temp/devel_0.patch

HS@NB2 /d/PDev/drupal/d6/sites/all/modules/devel (6.x-1.x)
$ git apply -R /d/temp/devel_0.patch

HS@NB2 /d/PDev/drupal/d6/sites/all/modules/devel (6.x-1.x)
$ git apply -p0 /d/temp/devel_0.patch

HS@NB2 /d/PDev/drupal/d6/sites/all/modules/devel (6.x-1.x)
$ git apply -R /d/temp/devel_0.patch

HS@NB2 /d/PDev/drupal/d6/sites/all/modules/devel (6.x-1.x)
$ git apply -p1 /d/temp/devel_0.patch
fatal: unable to find filename in patch at line 3

As far as I can tell

• it's a -p0 patch
• it applies without -p
• it applies with -p0
• it fails with -p1

I did try this before posting. That patch was rejected by the testbot because of the explicit -p1 (see the log in the OP above). It passes on my 1.7.4.msysgit.0 and it would probably have passed on the testbot without any -p.

Apparently, Git does a more sophisticated analysis. It won't accept its own -p0 patches, maybe because it realizes that the patch creator consciously deviated from the default, but it will accept those of other tools in order to not scare away the uninitiated.

Except for your drupal.demo_p0_patch.patch I cannot recall seeing any patch that I was unable to git apply (without -p) because of the -p0/-p1 issue. That's why I've never been able to understand what this fuss is all about, and I still don't understand why we need to specify -p1 on the testbot, since it's the default anyway. Why not just drop it and see what happens?

It can't get worse, can it? And maybe it'll get a whole lot better...

You're not listening to me, Randy.

I do not intend to use anything other than Git. But we have users like the OP in #1345240: Patch not to break xhprof reporting when site name contains spaces who are so scared by our efforts to train them to use -p1 and never ever -p0 that they barely dare to breathe when they submit their first patch, and they feel like they have to apologize in advance for the tool they prefer to use:

(This is the first patch I've submitted so I'd appreciate knowing if I've done it the right way. I don't know how to use git so I generated this from my Eclipse IDE using SVN.)

If you removed the -p1 switch from the testbot and let Git default to whatever it wants, then you could stop training and scaring people, and it would just work.

One thing I notice is that your patch was *not* created with git.

So what?

If I apply it and then create a patch with git, it will *not* apply here.

You make it sound like it were contagious... It's not. It's just a diff in your file. You git diff --no-prefix — you get a patch that requires -p0 to apply. That has nothing to do with where the diff came from.

That's what I wrote in #7: "Git won't accept its own -p0 patches [without -p0], maybe because it realizes that the patch creator consciously deviated from the default [by giving --no-prefix], but it will accept those [prefix-less patches] of other tools in order to not scare away the uninitiated [who don't use Git and don't understand what we're talking about]."

Your experiment confirms exactly what I wrote. That's no reason to be scared, or to tell people they have to use other tools than the ones they're comfortable with, or to use switches on Git that make it less friendly than it wants to be, unless we have a good reason to do so. It seems that we don't.