27

I would like to create a patch from a specific gcc branch comparing it to the official releases; so when I unpack the tarball from the stable release, I can apply the patch and get the equivalent of what was in that specific branch .

It's the first time I need to create a patch, so it's my very first time doing this and my main concern is to get the options and the parsing right since we are talking about an extremely important piece of software 

diff -crB GccStable GccGit > /tmp/fromStabletoBranch.patch

Is this enough and the best way of doing it ?

user2485710
  • 1,403
  • 4
  • 15
  • 26
  • The usual good practices here involve version control or some variant of these. This includes, mercurial, git, and their associated patch queue extensions. You could also consider quilt. Perhaps you could go into more detail as to what you are trying to do? – Faheem Mitha Oct 14 '14 at 21:46
  • @FaheemMitha what do you mean with "more details" ? I have a version of `gcc` from the official stable `tar.bz2` and another unstable version of it from a `git` repository, I would like to create a patch, of course I would like to compare just against the `master` branch, not the entire repository . – user2485710 Oct 14 '14 at 21:50
  • OK, well, sure you can use something as simple as diff. but using version control is generally preferable. For one thing, it makes it much harder to lose track of what you are doing. – Faheem Mitha Oct 14 '14 at 21:57
  • @FaheemMitha I don't understand what you are suggesting, my `tar.bz2` is clearly not a `git` repository, how do you think I should proceed ? – user2485710 Oct 14 '14 at 22:01
  • Well, you could (a) clone the relevant version as a git repo from upstream, or (b) put the source (or part of it) under version control yourself. And you don't have to use git. You could use mercurial, or even svn (though I don't recommend it). Personally I use mercurial and like it. – Faheem Mitha Oct 14 '14 at 22:05
  • @FaheemMitha you are basically suggesting a local `merge` ? Anyway I don't this will cut it in my case, I want something that is as easier as possible to apply, plus I don't need yet another repository, I can't screw up a patch, a repository is way too cumbersome for such task and much easier to get wrong especially over time and usage . – user2485710 Oct 14 '14 at 22:21
  • I never said anything about a merge, and a repository is not cumbersome in the slightest. You do know what version control is, right? – Faheem Mitha Oct 15 '14 at 08:15
  • @FaheemMitha please write a post with a complete explanation, this is getting confusing to say the least . Yes I know what versioning control software does . – user2485710 Oct 15 '14 at 08:29
  • 1
    Do a search for "creating patches using version control". Two previous answers I've written which are related are http://unix.stackexchange.com/a/127810/ and http://unix.stackexchange.com/a/139817/ – Faheem Mitha Oct 15 '14 at 08:43

2 Answers2

40

Yes, this is a good way to create a patch.

In short:

  1. To create patch for single file your command may look like

    diff -Naru file_original file_updated > file.patch

    where

    • -N: treat absent files as empty
    • -a: treat all files as text
    • -r: recursively compare any subdirectories found
    • -u: output NUM (default 3) lines of unified context

  2. To create patch for whole directory:

    diff -crB dir_original dir_updated > dfile.patch

    where

    • -c: output NUM (default 3) lines of copied context
    • -r: recursively compare any subdirectories
    • -B: ignore changes whose lines are all blank

After all to apply this patch one can run

patch -p1 --dry-run < dfile.patch

where switch p instructs patch to strip the path prefix so that files will be identified correctly. In most cases it should be 1.

Remove --dry-run if you are happy from the result printed on the screen.

jimmij
  • 46,064
  • 19
  • 123
  • 136
  • question: what is supposed to happen if a directory or a file gets deleted in the `dir_updated` compared to what there was in `dir_original` ? The `diff` takes care of that too or it gets skipped ? – user2485710 Oct 15 '14 at 07:47
  • @user2485710 diff sees which files were deleted. `file.patch` in just a text file, so you can open it in any editor or just `cat` it and you will see a line like `only in dir_original: missingfile.txt` – jimmij Oct 15 '14 at 14:51
  • ok, but then `patch` will delete that `missingfile.txt` or what else ? – user2485710 Oct 15 '14 at 15:14
  • It depends. If you want to remove them use `diff -N ...` as in my first example. Normally `patch` will remove empty files by default. If you don't want, just use only `diff -crB` as in your question. Also in some (rare) cases `-E` option in `patch` command is needed to remove empty files, after patch manual: `if the input is not a context diff or if patch is conforming to POSIX, patch does not remove empty patched files unless this option is given` – jimmij Oct 15 '14 at 15:33
  • i've run `diff -Naru dir/file dir/file.new > diff.patch` getting `can't find file to patch at input line 3` with `patch -p0 --dry-run < diff.patch` first line of patch reads `--- dir/file` second one `+++ dir/file.new` with timestamps, the diff seems to be just right, is there any option how report the exact filenames the command looks for? – ptica Mar 14 '19 at 11:57
  • Using `--dry-run` **will not actually change any files!** It "Print the results of applying the patches without actually changing any files." – Instein Jun 17 '22 at 22:55
0

If you want to compare latest git checkin to some stable version, just go git diff the-stable-version (will need to figure out what tag describes it, probably the exact version number or some variant) in the repository. git keeps the full history of the project (normally, there are ways to grab just a part). It doesn't matter if the-stable-version is in some different development branch (i.e., the development forked off and the stable branch got some last-minute fixes).

vonbrand
  • 18,156
  • 2
  • 37
  • 59