path: root/open_issues/git-core-2.mdwn

[[!meta copyright="Copyright © 2008, 2009, 2010, 2011, 2013, 2016 Free Software
Foundation, Inc."]]

[[!meta license="""[[!toggle id="license" text="GFDL 1.2+"]][[!toggleable
id="license" text="Permission is granted to copy, distribute and/or modify this
document under the terms of the GNU Free Documentation License, Version 1.2 or
any later version published by the Free Software Foundation; with no Invariant
Sections, no Front-Cover Texts, and no Back-Cover Texts.  A copy of the license
is included in the section entitled [[GNU Free Documentation
License|/fdl]]."]]"""]]

[[!meta title="Hiccups of Git"]]

[[!tag open_issue_porting]]

[[!toc]]


# 2008-12

On the otherwise-idle flubber:

    $ git clone git://sources.redhat.com/git/glibc.git
    Initialized empty Git repository in /media/data/home/tschwinge/tmp/glibc/glibc/.git/
    remote: Generating pack...
    remote: Done counting 380933 objects.
    remote: Deltifying 380933 objects...
    remote:  100% (380933/380933) done
    remote: Total 380933 (delta 294166), reused 380686 (delta 294002)
    Receiving objects: 100% (380933/380933), 70.31 MiB | 27 KiB/s, done.
    Resolving deltas: 100% (294166/294166), done.
    error: git-checkout-index: unable to create file iconvdata/ibm1122.c (Interrupted system call)
    error: git-checkout-index: unable to create file localedata/charmaps/IBM862 (Interrupted system call)
    Checking out files: 100% (10676/10676), done.
    $ git status
    # On branch master
    # Changed but not updated:
    #   (use "git add <file>..." to update what will be committed)
    #
    #       modified:   iconvdata/ibm1122.c
    #       modified:   localedata/charmaps/IBM862
    #
    no changes added to commit (use "git add" and/or "git commit -a")
    $ ls -l iconvdata/ibm1122.c localedata/charmaps/IBM862
    -rw-r--r-- 1 tschwinge tschwinge 0 2008-12-15 15:49 iconvdata/ibm1122.c
    -rw-r--r-- 1 tschwinge tschwinge 0 2008-12-15 15:49 localedata/charmaps/IBM862

So these files are indeed of zero-length in the checked-out tree.  Is this
Git's fault or something else's?

Fixing this situation is easy enough:

    $ git checkout -- iconvdata/ibm1122.c localedata/charmaps/IBM862
    $ git status
    # On branch master
    nothing to commit (working directory clean)


## 2010-03-16

Still seen.


## IRC, freenode, #hurd, 2013-10-10

    <sea`> Huh? I've cloned the 'hurd' repository and I'm attempting to compile
      it, but the 'rtnetlink.h' header in
      'hurd/pfinet/linux-src/include/linux/' is just blank. (Which leads to an
      error later down when a macro that's supposed to be defined in there is
      first used)
    <sea`> So I'm just wondering, is that file really blank? Or is this some
      unexpected error of decompression? 
    <braunr> clone again and see
    <braunr> the file is definitely not empty
    <sea`> I cloned it twice, both have that file blank. BUT, I want to point
      out that both clones do have some decompression errors. (Some files are
      missing chunks in /both/ cloned repositories).
    <braunr> where did you clone it from ?
    <sea`> git.sv.gnu.org/hurd/hurd.git
    <braunr> hum decompression errors ?
    <braunr> can you paste them please ?
    <sea`> Hmm, I can clone again and show you an example if I find one
    <sea`> This was on the hurd. When I run: git clone $repo;, it seems to fail
      almost randomly with "incorrect header check", but when it does succeed,
      occasionally some files are missing chunks
    <sea`> and apparently entire files can be blank
    <braunr> http or git ?
    <sea`> git.
    <braunr> that's really weird
    <braunr> actually i don't even have problems with http any more nowadays ..
    <sea`> This is using the hurd image from sthibault
    <sea`> So once I get it recompiled and shuffle in the new binaries, the
      problem should probably go away
    <braunr> no
    <braunr> well maybe but
    <braunr> don't recompile
    <braunr> upgrade packages instead
    <sea`> Alright, I'll do an upgrade instead. Why that path specifically?
    <braunr> rebuilding is long
    <braunr> i wonder if the image you got is corrupted
    <braunr> compute the checksum
    <braunr> we've had weird reports in the past about the images he provides
    <braunr> well not the images themselves, but differences after dowloading
      ..
    <braunr> downloading*
    <sea`> The MD5SUMS file on his site isn't including the values for the most
      recent images.
    <sea`> It stops at 2012-12-28
    <braunr> hummm
    <sea`> Anyway, let's see. git clone failed again:
    <sea`> Receiving objects: 100% (50955/50955), 15.48 MiB | 42 KiB/s, done.
    <sea`> error: inflate: date stream error (incorrect header check) <- This
      is the interesting part
    <sea`> fatal: serious inflate inconsistency
    <sea`> fatal: index-pack failed
    <braunr> not intereseting enough unfortunately
    <braunr> but it might come from savannah too

    <sea`> Ehm. The whole thing locked up badly. I'll reboot it and try again.
    <braunr> are you sure it locked oO ?
    <braunr> the hurd can easily become unresponsive when performing io
      operations
    <braunr> but you need more than such a git repository to reach that state
    <sea`> Yeah, that happens occasionally. It's not related to git, but rather
      it happens when I cancel some command.
    <braunr> your image must be corrupted
    <braunr> have you enabled host io caching btw ?
    <sea`> By now it's corrupted for sure..everytime it crashes the filesystem
      gets into a weird state. 
    <sea`> I'll unpack a fresh image, then update the packages, and then try
      cloning this git repository.
    <braunr> i'll get the image too so we can compare sums
    <sea`> 957bb0768c9558564f0c3e0adb9b317e  ./debian-hurd.img.tar.gz
    <sea`> Which unpacks to: debian-hurd-20130504.img
    <azeem_> the NSA might backdoor the Hurd, in anticipation of our scheduled
      world-dominance
    <braunr> for now they're doing it passively :
    <braunr> :p
    <braunr> sea`: same thing here
    <braunr> sea`: if you still have problems, the image itself might be wrong
    <braunr> in which case you should try with the debian network installer
    <sea`> Ah, so if problems persist, try with the network installer. Okay
    <sea`> Is there some recipe for constructing a hurd/mach minimal
      environment? 
    <sea`> A system with only just enough tools and libraries to compile and
      poke at things.
    <braunr> not currently
    <braunr> we all work in debian environments
    <braunr> the reason being that a lot of patches are queued for integration
      upstream


# 2010-11-17

A very similar issue.  The working tree had a lot of
differences to HEAD.

    tschwinge@grubber:~/tmp/gcc/hurd $ git reset --hard HEAD
    error: unable to unlink old 'gcc/config/darwin.h' (Interrupted system call)
    Checking out files: 100% (1149/1149), done.
    fatal: Could not reset index file to revision 'HEAD'.
    tschwinge@grubber:~/tmp/gcc/hurd $ git reset --hard HEAD
    error: unable to unlink old 'gcc/config/iq2000/iq2000.md' (Interrupted system call)
    error: git checkout-index: unable to create file gcc/config/lm32/lm32.c (File exists)
    Checking out files: 100% (1149/1149), done.
    fatal: Could not reset index file to revision 'HEAD'.
    tschwinge@grubber:~/tmp/gcc/hurd $ ls -l gcc/config/iq2000/iq2000.md gcc/config/lm32/lm32.c
    ls: cannot access gcc/config/iq2000/iq2000.md: No such file or directory
    -rw-r--r-- 1 tschwinge tschwinge 32159 Nov 17 19:09 gcc/config/lm32/lm32.c
    tschwinge@grubber:~/tmp/gcc/hurd $ git reset --hard HEAD
    error: git checkout-index: unable to create file gcc/fortran/expr.c (Interrupted system call)
    Checking out files: 100% (1149/1149), done.
    fatal: Could not reset index file to revision 'HEAD'.
    tschwinge@grubber:~/tmp/gcc/hurd $ git reset --hard HEAD
    error: git checkout-index: unable to create file gcc/config/sol2.h (Interrupted system call)
    Checking out files: 100% (1149/1149), done.
    fatal: Could not reset index file to revision 'HEAD'.
    tschwinge@grubber:~/tmp/gcc/hurd $ git reset --hard HEAD
    error: unable to unlink old 'gcc/config/i386/i386.c' (Interrupted system call)
    Checking out files: 100% (1149/1149), done.
    fatal: Could not reset index file to revision 'HEAD'.
    tschwinge@grubber:~/tmp/gcc/hurd $ git reset --hard HEAD
    Checking out files: 100% (1149/1149), done.
    HEAD is now at fe3e43c Merge commit 'refs/top-bases/hurd/master' into hurd/master


## 2010-12-22

grubber:

    $ git remote update
    Fetching savannah
    remote: Counting objects: 582331, done.
    remote: Compressing objects: 100% (124133/124133), done.
    remote: Total 582331 (delta 460856), reused 578352 (delta 457598)
    Receiving objects: 100% (582331/582331), 525.15 MiB | 204 KiB/s, done.
    fatal: cannot pread pack file: Interrupted system call
    fatal: index-pack failed
    error: Could not fetch savannah


## 2011-06-10

coulomb.SCHWINGE, checking out [[binutils]]' master branch,
starting from an empty working directory (after an external `git push`):

    $ git checkout -f
    fatal: cannot create directory at 'gas/testsuite/gas/bfin': Interrupted system call
    $ git checkout -f
    error: unable to create file gas/testsuite/gas/i386/ilp32/x86-64-sse4_1-intel.d (File exists)
    warning: unable to unlink gas/testsuite/gas/m68k-coff: Operation not permitted
    fatal: cannot create directory at 'gas/testsuite/gas/m68k-coff': Operation not permitted
    $ git checkout -f
    error: unable to create file gas/testsuite/gas/h8300/h8300.exp (File exists)
    error: unable to create file gas/testsuite/gas/i386/x86-64-addr32-intel.d (File exists)
    error: unable to create file gas/testsuite/gas/ia64/secname.d (File exists)
    error: unable to create file gas/testsuite/gas/m68k/pr11676.s (File exists)
    Checking out files: 100% (12315/12315), done.
    $ git status
    # On branch master
    # Changes not staged for commit:
    #   (use "git add <file>..." to update what will be committed)
    #   (use "git checkout -- <file>..." to discard changes in working directory)
    #
    #       modified:   gas/testsuite/gas/h8300/h8300.exp
    #       modified:   gas/testsuite/gas/i386/x86-64-addr32-intel.d
    #       modified:   gas/testsuite/gas/ia64/secname.d
    #       modified:   gas/testsuite/gas/m68k/pr11676.s
    #
    no changes added to commit (use "git add" and/or "git commit -a")
    $ rm gas/testsuite/gas/h8300/h8300.exp gas/testsuite/gas/i386/x86-64-addr32-intel.d gas/testsuite/gas/ia64/secname.d gas/testsuite/gas/m68k/pr11676.s
    $ git checkout -f
    $ git status
    # On branch master
    nothing to commit (working directory clean)


# Analysis

## 2011-06-13

Running `git checkout -f` under GDB:

    error: git checkout-index: unable to create file gas/testsuite/gas/cris/string-1.s (File exists)
    error: git checkout-index: unable to create file gas/testsuite/gas/i386/x86-64-sse-check.d (File exists)
    error: git checkout-index: unable to create file gas/testsuite/gas/i386/x86-64-sse4_1.d (File exists)
    error: git checkout-index: unable to create file gas/testsuite/gas/ppc/astest.d (File exists)
    error: git checkout-index: unable to create file gas/testsuite/gas/tic6x/reloc-bad-4.s (File exists)
    warning: unable to unlink include/cgen: Operation not permitted
    fatal: cannot create directory at 'include/cgen': Operation not permitted

Again:

    error: git checkout-index: unable to create file gas/config/te-vxworks.h (File exists)
    error: git checkout-index: unable to create file gas/testsuite/gas/cris/string-1.s (File exists)
    error: git checkout-index: unable to create file gas/testsuite/gas/d10v/warning-019.s (File exists)
    error: git checkout-index: unable to create file gas/testsuite/gas/i860/dual03.s (File exists)
    error: git checkout-index: unable to create file ld/testsuite/ld-mmix/sec-7a.s (File exists)
    warning: unable to unlink ld/testsuite/ld-powerpc: Operation not permitted
    fatal: cannot create directory at 'ld/testsuite/ld-powerpc': Operation not permitted

And: [[git_duplicated_content]].

All these (very likely) have the same root cause: `SA_RESTART` restarting too
much.

With `git checkout`, Git uses in progress.c a SIGALRM handler (`SA_RESTART`;
invoked every second via `setitimer(ITIMER_REAL)`) to display status messages:
*x % already checked out*.

To avoid the status update signals every second, in
`[git]/progress.c:start_progress_delay` we can just return `NULL` (manually in
GDB, for example), then both the *error: git checkout-index* and the
[[duplicated content|git_duplicated_content]] issues go away.

I'm guessing that when returning from a `SA_RESTART` signal handler, too much
of the \`\`syscall''s is being restarted.  For example, if a file has already
been created, the restarted creation attempt would fail: *File exists*.  If
data has been written, it might get written again (duplication issue). Then,
there are cases where `unlink` apparently returns EINTR, which is not kosher
either.  Etc.

Do we have problems with `SA_RESTART` vs. the atomicity of our syscall-alikes?


## IRC, freenode, #hurd, 2013-01-30

    <braunr> hm, let's try to clone a huge repository
    <braunr> hm, cloning a whole linux repo, and still no problem :)
    <pinotree> weren't most/all the issues at unpack time?
    <braunr> i don't remember
    <braunr> we'll see when it gets there
    <braunr> the longest part is "resolving deltas", for which ext2fs is
      clearly the big bottleneck (no I/O, page-cache only, but still)
    <braunr> pinotree: well, slow, but no error


### IRC, freenode, #hurd, 2013-01-31

    <braunr> fyi, i've tried several checkouts of big repositories, and never
      got a single error
    <braunr> youpi: looks like the recent fixes also solved some git issues we
      had
    <braunr> i could clone big repositories without any problem
    <youpi> cool :)