What's happening to the index

Hi,

I'm using git version 1.8.3.1. I have a process for building RPMs in my repository. The RPMs are versioned using a "build number". In order to maintain uniqueness for this build number, I have stored this number to a file which exists on only a unique branch.

The build process, for an actual release, is intended to be done by branching on a tag. During that process, a BASH script is called with checks out "rpm", fetches the repo from remote, merges with origin/rpm, increments the number and pushes back to origin. It then returns the branch to the original branch in which the script was called. The make recipe looks like this:

release:
     make clean
     cd ../..  && ../tools/increlnum && cd -
     if [[ $(TAG) = ......... ]]; then \
         git checkout -b rpm_build_$(TAG) $(TAG); \
     fi
     make rpm RPM_BUILD_NUM=$(shell git show rpm:./rpm_build_num)

After the line calling increlnum is executed, I often have issues with make unable to spawn the next command because it can't read the current directory info. Make stops with errors and I'm done. I have the branch name displayed in my PS1 prompt. I've noticed that, when make errors at this point, the branch isn't displayed. It's as if the index has become unstable (or something similar). If I do: cd .. && cd -; all is well.

What is the problem (I'd really like to understand) and what might I do to correct it?

Andy

"Andy Falanga (afalanga)" <afalanga@micron.com> writes:

This may happen if you delete the current directory, even if your re-create it afterwards. For example:

/tmp/test$ rm -fr /tmp/test && mkdir /tmp/test /tmp/test$ touch foo touch: cannot touch ‘foo’: No such file or directory /tmp/test$ cd /tmp/test /tmp/test$ touch foo /tmp/test$

This is unrelated from Git, but maybe you asked Git to delete a directory (by switching to a branch which doesn't contain a directory for example).

This is a typical symptom of the issue above.

On 06/28/2016 09:17 AM, Matthieu Moy wrote:

Thank you for the insight: very interesting. After asking another colleague how he solved this issue, I've re-written my increlnum script to, instead of working within my working tree, clone a temporary of this one branch only. Then, it increments the number and pushes back to the origin. Once completed, the temporary clone is deleted.

The strange thing now is, after the script exits, I then call "git fetch" in the recipe. I can see from the output of make that the remote db is fetched. However, when I call "git show origin/rpm:path/to/rpm_build_num" from the makefile I get the *previous* number. Yet, as soon as the make process exits, I call "git show origin/rpm:path/to/rpm_build_num" and it shows the correct number! What gives? Is there some sort of strange file caching that happening when make starts that, although the local db is updated, I don't get what I'm after?

Andy

"Andy Falanga (afalanga)" <afalanga@micron.com> writes:

My bet would be that you are running the commands in two different repositories. Try running

  pwd && git show origin/rpm:path/to/rpm_build_num

In the Makefile and after make completes, and check that pwd returns the same thing. Also, to avoid getting the issue you previously had, try

  cd $(pwd) && pwd && git show origin/rpm:path/to/rpm_build_num

too.

Git can keep information either in RAM, hence not shared between git invocations (so running git within or outside the Makefile wouldn't matter), or on disk, but then inside or outside the Makefile doesn't matter either.

On 29 June 2016 at 08:49, Andy Falanga (afalanga) <afalanga@micron.com> wrote:

I don't have time to look at your git issue, but I write this quick note just in case it might help you to be aware that 'gnu make' apparently does implement internal directory caching.

This is known to cause unexpected results in makefiles whose recipes change the filesystem in ways that make does not notice.

See for example: http://lists.gnu.org/archive/html/help-make/2015-02/msg00012.html

and related bug reports, for example: https://savannah.gnu.org/bugs/index.php?41273

You might want to create a simple runnable example and ask on the extremely helpful gnu-make mailing list: https://lists.gnu.org/mailman/listinfo/help-make