# Possible minor bug in Git 10 messages from 2019-01-31 to 2019-02-10. Participants: Angelo Melonas, Torsten Bögershausen, Giuseppe Crinò, Johannes Schindelin, Philip Oakley, Giuseppe Crino'. Thread: https://gitlist.dev/t/50367 ## Angelo Melonas, 2019-01-31 07:01 Subject: Possible minor bug in Git Message-ID: URL: https://gitlist.dev/e/CAHYHVP%2Bty7i7a_abT-th_HR2X-8X6mLmpHPR_3VXv5LpWUsrhQ%40mail.gmail.com ``` Good day, I found a potential bug in Git for Windows. The bug can be reproduced as follows: 1. Modify a file so that it shows up as "untracked" when executing the "git status" command. 2. Attempt to "git add" the file, but change the case of a single letter. The command executes but no warning or error is displayed. 3. Execute "git status" again and it still shows the file as "untracked". Please let me know if I am mistaken. I also have a screenshot demonstrating the "bug" which I cannot attach to this email, but which can be requested. Have a great day. Kind regards, Angelo Melonas ``` ## Angelo Melonas, 2019-01-31 07:29 Subject: Re: Possible minor bug in Git Message-ID: URL: https://gitlist.dev/e/CAHYHVP%2BDvbq_aOMy1_Kq0LHJsd4r%2B4GiP3N3R4A9ouAoJ4gw9g%40mail.gmail.com In-Reply-To: ``` Just to clarify, I made a mistake above. Instead of "untracked", I meant "unstaged". I apologise for the confusion. On Thu, Jan 31, 2019 at 9:01 AM Angelo Melonas wrote: > > Good day, > > I found a potential bug in Git for Windows. The bug can be reproduced > as follows: > > 1. Modify a file so that it shows up as "untracked" when executing the > "git status" command. > 2. Attempt to "git add" the file, but change the case of a single > letter. The command executes but no warning or error is displayed. > 3. Execute "git status" again and it still shows the file as "untracked". > > Please let me know if I am mistaken. I also have a screenshot > demonstrating the "bug" which I cannot attach to this email, but which > can be requested. > > Have a great day. > > Kind regards, > Angelo Melonas ``` ## Torsten Bögershausen, 2019-01-31 20:34 Subject: Re: Possible minor bug in Git Message-ID: <20190131203447.aovalz2vzsjpdjdv@tb-raspi4> URL: https://gitlist.dev/e/20190131203447.aovalz2vzsjpdjdv%40tb-raspi4 In-Reply-To: ``` On Thu, Jan 31, 2019 at 09:29:28AM +0200, Angelo Melonas wrote: > Just to clarify, I made a mistake above. Instead of "untracked", I > meant "unstaged". > I apologise for the confusion. > > On Thu, Jan 31, 2019 at 9:01 AM Angelo Melonas wrote: > > > > Good day, > > > > I found a potential bug in Git for Windows. The bug can be reproduced > > as follows: > > > > 1. Modify a file so that it shows up as "untracked" when executing the > > "git status" command. > > 2. Attempt to "git add" the file, but change the case of a single > > letter. The command executes but no warning or error is displayed. > > 3. Execute "git status" again and it still shows the file as "untracked". > > > > Please let me know if I am mistaken. I also have a screenshot > > demonstrating the "bug" which I cannot attach to this email, but which > > can be requested. > > > > Have a great day. > > > > Kind regards, > > Angelo Melonas See the example below, Git tracks AA.txt, so you must run git add AA.txt After a commit, you can tell Git that the file was renamed: git mv AA.txt Aa.txt (and then a commit) ------------------------------ user@mac:/tmp/tt2> git init Initialized empty Git repository in /private/tmp/tt2/.git/ user@mac:/tmp/tt2> echo AAA > AA.txt user@mac:/tmp/tt2> git add AA.txt user@mac:/tmp/tt2> git commit -m AA.txt [master (root-commit) f102760] AA.txt 1 file changed, 1 insertion(+) create mode 100644 AA.txt user@mac:/tmp/tt2> echo BB >AA.txt user@mac:/tmp/tt2> mv AA.txt Aa.txt user@mac:/tmp/tt2> git status On branch master Changes not staged for commit: (use "git add ..." to update what will be committed) (use "git checkout -- ..." to discard changes in working directory) modified: AA.txt no changes added to commit (use "git add" and/or "git commit -a") user@mac:/tmp/tt2> git add AA.txt user@mac:/tmp/tt2> git status On branch master Changes to be committed: (use "git reset HEAD ..." to unstage) modified: AA.txt user@mac:/tmp/tt2> ``` ## Angelo Melonas, 2019-02-01 08:02 Subject: Re: Possible minor bug in Git Message-ID: URL: https://gitlist.dev/e/CAHYHVPJdbEOPaAoyAhOtJ%2BwkeXZdowAVqe1Tp_%3DLOKzuuzA%2BwQ%40mail.gmail.com In-Reply-To: <20190131203447.aovalz2vzsjpdjdv@tb-raspi4> ``` Hi Torsten, Thank you so much for getting back to me. Unfortunately, I believe there is a misunderstanding, as I may have explained what I found to be a possible bug incorrectly. The file that is originally added (and then later modified) is never renamed or moved. As you will see below, when a user attempts to add a file, but with the incorrect case, the Git CLI responds the same way it would if a file was correctly added (i.e., by displaying nothing). However, in the above case, when you enter "git status", you find that the file was never actually added, and remains unstaged. A possible solution to this can be a simple error message similar to attempting to add a file with its name misspelt. Using your example, I have illustrated this in the text below: C:\Example>git init Initialized empty Git repository in C:/Example/.git/ C:\Example>echo AAA > AA.txt C:\Example>git add AA.txt C:\Example>git commit -m AA.txt [master (root-commit) d550af0] AA.txt 1 file changed, 1 insertion(+) create mode 100644 AA.txt C:\Example>echo BB > AA.txt C:\Example>git status On branch master Changes not staged for commit: (use "git add ..." to update what will be committed) (use "git checkout -- ..." to discard changes in working directory) modified: AA.txt no changes added to commit (use "git add" and/or "git commit -a") C:\Example>git add Aa.txt C:\Example>git status On branch master Changes not staged for commit: (use "git add ..." to update what will be committed) (use "git checkout -- ..." to discard changes in working directory) modified: AA.txt no changes added to commit (use "git add" and/or "git commit -a") C:\Example>git add A.txt fatal: pathspec 'A.txt' did not match any files C:\Example>git add AA.txt C:\Example>git status On branch master Changes to be committed: (use "git reset HEAD ..." to unstage) modified: AA.txt I am looking forward to hearing from you again. Kind regards, Angelo Melonas On Thu, Jan 31, 2019 at 10:34 PM Torsten Bögershausen wrote: > > On Thu, Jan 31, 2019 at 09:29:28AM +0200, Angelo Melonas wrote: > > Just to clarify, I made a mistake above. Instead of "untracked", I > > meant "unstaged". > > I apologise for the confusion. > > > > On Thu, Jan 31, 2019 at 9:01 AM Angelo Melonas wrote: > > > > > > Good day, > > > > > > I found a potential bug in Git for Windows. The bug can be reproduced > > > as follows: > > > > > > 1. Modify a file so that it shows up as "untracked" when executing the > > > "git status" command. > > > 2. Attempt to "git add" the file, but change the case of a single > > > letter. The command executes but no warning or error is displayed. > > > 3. Execute "git status" again and it still shows the file as "untracked". > > > > > > Please let me know if I am mistaken. I also have a screenshot > > > demonstrating the "bug" which I cannot attach to this email, but which > > > can be requested. > > > > > > Have a great day. > > > > > > Kind regards, > > > Angelo Melonas > > See the example below, Git tracks AA.txt, so you must run > git add AA.txt > > After a commit, you can tell Git that the file was renamed: > git mv AA.txt Aa.txt > (and then a commit) > > > > ------------------------------ > > user@mac:/tmp/tt2> git init > Initialized empty Git repository in /private/tmp/tt2/.git/ > user@mac:/tmp/tt2> echo AAA > AA.txt > user@mac:/tmp/tt2> git add AA.txt > user@mac:/tmp/tt2> git commit -m AA.txt > [master (root-commit) f102760] AA.txt > 1 file changed, 1 insertion(+) > create mode 100644 AA.txt > user@mac:/tmp/tt2> echo BB >AA.txt > user@mac:/tmp/tt2> mv AA.txt Aa.txt > user@mac:/tmp/tt2> git status > On branch master > Changes not staged for commit: > (use "git add ..." to update what will be committed) > (use "git checkout -- ..." to discard changes in working directory) > > modified: AA.txt > > no changes added to commit (use "git add" and/or "git commit -a") > user@mac:/tmp/tt2> git add AA.txt > user@mac:/tmp/tt2> git status > On branch master > Changes to be committed: > (use "git reset HEAD ..." to unstage) > > modified: AA.txt > > user@mac:/tmp/tt2> ``` ## Torsten Bögershausen, 2019-02-02 06:38 Subject: Re: Possible minor bug in Git Message-ID: <20190202063828.4kjtdmrsm7g4eyg3@tb-raspi4> URL: https://gitlist.dev/e/20190202063828.4kjtdmrsm7g4eyg3%40tb-raspi4 In-Reply-To: ``` On Fri, Feb 01, 2019 at 10:02:50AM +0200, Angelo Melonas wrote: > Hi Torsten, > > Thank you so much for getting back to me. > > Unfortunately, I believe there is a misunderstanding, as I may have > explained what I found to be a possible bug incorrectly. > The file that is originally added (and then later modified) is never > renamed or moved. > As you will see below, when a user attempts to add a file, but with > the incorrect case, the Git CLI responds the same way it would if a > file was correctly added (i.e., by displaying nothing). > However, in the above case, when you enter "git status", you find that > the file was never actually added, and remains unstaged. > A possible solution to this can be a simple error message similar to > attempting to add a file with its name misspelt. > > Using your example, I have illustrated this in the text below: > > C:\Example>git init > Initialized empty Git repository in C:/Example/.git/ > > C:\Example>echo AAA > AA.txt > > C:\Example>git add AA.txt > > C:\Example>git commit -m AA.txt > [master (root-commit) d550af0] AA.txt > 1 file changed, 1 insertion(+) > create mode 100644 AA.txt > > C:\Example>echo BB > AA.txt > > C:\Example>git status > On branch master > Changes not staged for commit: > (use "git add ..." to update what will be committed) > (use "git checkout -- ..." to discard changes in working directory) > > modified: AA.txt > > no changes added to commit (use "git add" and/or "git commit -a") > > C:\Example>git add Aa.txt > > C:\Example>git status > On branch master > Changes not staged for commit: > (use "git add ..." to update what will be committed) > (use "git checkout -- ..." to discard changes in working directory) > > modified: AA.txt > > no changes added to commit (use "git add" and/or "git commit -a") > > C:\Example>git add A.txt > fatal: pathspec 'A.txt' did not match any files > > C:\Example>git add AA.txt > > C:\Example>git status > On branch master > Changes to be committed: > (use "git reset HEAD ..." to unstage) > > modified: AA.txt > > I am looking forward to hearing from you again. > > Kind regards, > Angelo Melonas > Actually yes, you can call this a minor bug. Does anybody want to work on it ? May be as a first-time-Git contribution ? ``` ## Giuseppe Crinò, 2019-02-06 22:17 Subject: Re: Possible minor bug in Git Message-ID: <20190206221759.338-1-g@evil.localdomain> URL: https://gitlist.dev/e/20190206221759.338-1-g%40evil.localdomain In-Reply-To: <20190202063828.4kjtdmrsm7g4eyg3@tb-raspi4> ``` I wanted to have a look at the bug, and I can correctly reproduce it using version 2.20.1.windows.1. To start to even think of fixing this bug I need to build the source for Windows, but I got lost on how to do that. Is it correct that I should cross-compile from a POSIX system (GNU/Linux), using x86_64-w64-mingw32-gcc and Gnulib to produce a static executable? Am I missing something? How does people here build for Windows? Giuseppe ``` ## Johannes Schindelin, 2019-02-07 19:41 Subject: Re: Possible minor bug in Git Message-ID: URL: https://gitlist.dev/e/nycvar.QRO.7.76.6.1902072040360.41%40tvgsbejvaqbjf.bet In-Reply-To: <20190206221759.338-1-g@evil.localdomain> ``` Hi Giuseppe, On Wed, 6 Feb 2019, Giuseppe Crinò wrote: > I wanted to have a look at the bug, and I can correctly reproduce it using version 2.20.1.windows.1. > > To start to even think of fixing this bug I need to build the source for Windows, but I got lost on how to do that. Does this help? https://github.com/git-for-windows/git/wiki/Building-Git Ciao, Johannes > > Is it correct that I should cross-compile from a POSIX system (GNU/Linux), using x86_64-w64-mingw32-gcc and Gnulib to produce a static executable? > > Am I missing something? How does people here build for Windows? > > Giuseppe > ``` ## Torsten Bögershausen, 2019-02-08 17:43 Subject: Re: Possible minor bug in Git Message-ID: <20190208174315.nnljdpbh4nyoxgd7@tb-raspi4> URL: https://gitlist.dev/e/20190208174315.nnljdpbh4nyoxgd7%40tb-raspi4 In-Reply-To: ``` On Fri, Feb 08, 2019 at 04:18:23PM +0100, Giuseppe Crinò wrote: > OK, I successfully built git on Windows (thanks Johannes!) and I'm now able to run it. > > As of 9f16cdd I can successfully reproduce the bug. > > Interestingly enough, I can reproduce the bug even for /usr/bin/git running inside Windows Subsystem for Linux. Part of the reason might be that both relies on the same lstat() call... (Note: `stat` inside the WSL is case _insensitive_). > > Now: what is the expected result for git running inside Windows? Should it die saying "fatal: pathspec ... did not match any files"? > > If that's the case, is the following a valid test case? > > diff --git a/t/t3700-add.sh b/t/t3700-add.sh > index 8ee4fc70ad..fadd7c74f6 100755 > --- a/t/t3700-add.sh > +++ b/t/t3700-add.sh > @@ -61,6 +61,11 @@ test_expect_success 'git add: filemode=0 should not get confused by symlink' ' > test_mode_in_index 120000 xfoo2 > ' > > +test_expect_success 'git add: pathspec is case-sensitive' ' > + echo new > file && > + test_must_fail git add File > +' > + In general, yes. There are 2 comments: This the "echo" line should have no ' ' after the '>': echo new >file && The other question is, if we should move that test case into t0050-filesystem.sh, but that is a matter of taste. diff --git a/t/t0050-filesystem.sh b/t/t0050-filesystem.sh index 192c94eccd..b8d6bad97a 100755 --- a/t/t0050-filesystem.sh +++ b/t/t0050-filesystem.sh @@ -106,6 +106,11 @@ test_expect_failure CASE_INSENSITIVE_FS 'add (with different case)' ' test "z$(git cat-file blob :$camel)" = z1 ' +test_expect_success CASE_INSENSITIVE_FS 'add (with wrong case)' ' + git reset --hard initial && + test_must_fail git add CAMELCASE +' + ``` ## Philip Oakley, 2019-02-09 18:19 Subject: Re: Possible minor bug in Git Message-ID: <788eefef-f677-f0ce-7e02-dd85b11da78a@iee.org> URL: https://gitlist.dev/e/788eefef-f677-f0ce-7e02-dd85b11da78a%40iee.org In-Reply-To: ``` Hi Guiseppe, On 07/02/2019 19:41, Johannes Schindelin wrote: > Hi Giuseppe, > > On Wed, 6 Feb 2019, Giuseppe Crinò wrote: > >> I wanted to have a look at the bug, and I can correctly reproduce it using version 2.20.1.windows.1. Thank you for having a look. The root cause of the issues will most probably be use of a case insensitive file system on Windows (and Mac). There is a configuration flag `core.ignoreCase` [1] that is normally auto detected that can be used to decide when the checks should be done and advice [2] or warnings given. There are also similar case issues with branch names should you want to go that far. In any case you should probably at least cover the full utf-8 filenames, not just ascii ones. >> >> To start to even think of fixing this bug I need to build the source for Windows, but I got lost on how to do that. > Does this help? > > https://github.com/git-for-windows/git/wiki/Building-Git > > Ciao, > Johannes > >> Is it correct that I should cross-compile from a POSIX system (GNU/Linux), using x86_64-w64-mingw32-gcc and Gnulib to produce a static executable? >> >> Am I missing something? How does people here build for Windows? >> >> Giuseppe -- Philip [1] https://git-scm.com/docs/git-config#git-config-coreignoreCase [2] https://git-scm.com/docs/git-config#git-config-advice ``` ## Giuseppe Crino', 2019-02-10 22:46 Subject: Re: Possible minor bug in Git Message-ID: <20190210224607.GA118@evil.localdomain> URL: https://gitlist.dev/e/20190210224607.GA118%40evil.localdomain In-Reply-To: <788eefef-f677-f0ce-7e02-dd85b11da78a@iee.org> ``` Setting `true` as the default for GIT_ICASE_PATHSPECS_ENVIRONMENT, when git is built on a Windows system, solves the bug. diff --git a/pathspec.c b/pathspec.c index 12c2b322b3..906cf24e3e 100644 --- a/pathspec.c +++ b/pathspec.c @@ -237,7 +237,11 @@ static inline int get_icase_global(void) static int icase = -1; if (icase < 0) + #if defined(GIT_WINDOWS_NATIVE) || defined(__CYGWIN__) + icase = git_env_bool(GIT_ICASE_PATHSPECS_ENVIRONMENT, 1); + #else icase = git_env_bool(GIT_ICASE_PATHSPECS_ENVIRONMENT, 0); + #endif return icase; } Unfortunately that fix introduces a regression too, tested in t/t3700-add.sh --- `error out when attempting to add ignored ones but add others`. I already spent some time to understand why, but got no luck: I have to dive deeper into the source code. In case I can fix the regression, is changing the default value of that env variable a good solution? Should I change the approach? Like leveraging core.ignorecase somewhere ...? On Sat, Feb 09, 2019 at 06:19:11PM +0000, Philip Oakley wrote: > The root cause of the issues will most probably be use of a case insensitive > file system on Windows (and Mac). There is a configuration flag > `core.ignoreCase` [1] that is normally auto detected that can be used to > decide when the checks should be done and advice [2] or warnings given. Thanks, Giuseppe ```