Permission denied on windows

[atsju]

Ninja 1.8.2 and 1.9.0 reproduced.
In a project creating many static libraries with cmake,

[build] FAILED: Common/05_MCU_abstract_layer/libmal_interrupt.a 
[build] cmd.exe /C "cd . && C:\GIT\FW_Toolchain\cmake\bin\cmake.exe -E remove Common\05_MCU_abstract_layer\libmal_interrupt.a && C:\GIT\FW_TOO~1\GNUTOO~1\7-2018~1\bin\ARM-NO~2.EXE qc Common\05_MCU_abstract_layer\libmal_interrupt.a  Common/05_MCU_abstract_layer/CMakeFiles/mal_interrupt.dir/interrupt/mal_interrupt.c.obj && C:\GIT\FW_TOO~1\GNUTOO~1\7-2018~1\bin\AR17F9~1.EXE Common\05_MCU_abstract_layer\libmal_interrupt.a && cd ."
[build] C:\GIT\FW_TOO~1\GNUTOO~1\7-2018~1\bin\AR17F9~1.EXE: Common\05_MCU_abstract_layer\libmal_interrupt.a: Permission denied

The issue is spurious, not always on same file. When running serial instead parallel, the problem disappear. Also just continuing build when it fails seem to permit finishing the build.
I found someone with the same problem (in german) here .
I think this could come from parallel (or too fast) access to same files in windows.

[atsju]

I reproduced on several windows 10 environments. I suspected mcafee on-scann access but this is not the case as one of the computer has no mcafee or other specific virus scan installed.
Please let me know if I must give some more information

Also in the project, many static libs depend on the same static lib.

[nicolasdespres]

Hi Julien, I think I have experienced the same problem but with precompiled headers on Window 10. Concurrent access to the same precompiled header resulted in permission denied error. Problem disappeared when running in serial instead of parallel. Cheers, Nico

…

On Mon, Sep 9, 2019 at 2:58 PM Julien Staub ***@***.***> wrote: I reproduced on several windows 10 environments. I suspected mcafee on-scann access but this is not the case as one of the computer has no mcafee or other specific virus scan installed. Please let me know if I must give some more information — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1646?email_source=notifications&email_token=AAEWBZ6ACAJR5FN4WPH4HFDQIZB6DA5CNFSM4IUIGKN2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD6HO6HQ#issuecomment-529461022>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAEWBZZFY65L4K32CPGOEEDQIZB6DANCNFSM4IUIGKNQ> .

-- Nicolas Desprès

[atsju]

Thank you Nicolas.
I used the same workaround but this should be fixed. I cannot share my example but do you have a shareable minimal code to reproduce the issue ? This could help to fix it.

[nicolasdespres]

Unfortunately not.

…

On Thu, Oct 31, 2019 at 9:16 AM Julien Staub ***@***.***> wrote: I used the same workaround but this should be fixed. I cannot share my example but do you have a shareable minimal code to reproduce the issue ? This could help to fix it. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1646?email_source=notifications&email_token=AAEWBZ5AQEON72HEYOOZBCDQRKH7VA5CNFSM4IUIGKN2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOECW4RCQ#issuecomment-548259978>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAEWBZ52FFFIWJNQTUFDDELQRKH7VANCNFSM4IUIGKNQ> .

-- Nicolas Desprès

[atsju]

A minimal project to reproduce the bug can be found in this repository. The project has mainly empty file but many diamond dependencies between static libraries.
I build on windows 10 with arm-none-eabi-gcc 7-2018-q2 on VScode.
The project may need to build several times to reproduce the issue.
Please let me know if you can reproduce and if you need more information.

[kromancer]

Maybe this is relevant, it might not be a ninja issue afterall:
https://bugs.launchpad.net/gcc-arm-embedded/+bug/1848002

[Vox-Ben]

I have this same problem, Ninja 1.10.0. Any news on a fix? Working around it by disabling threading is painful...

[leoatinnovation]

Hi Vox-Ben,

I'm also facing the same problem. what is the workaround ?

[Vox-Ben]

@leoatinnovation If you pass -j 1 to Ninja it runs only on a single thread (I think the default is three, in normal operation you can set it to whatever you like - ideally probably the same number of (virtual) processor cores that your build machine has). This avoids the bug by not running multiple threads that can conflict, but it reduces the speed. I got about a 50% speed reduction over the three-thread default, if you were otherwise using more threads to start with it'd be worse.

[Vox-Ben]

Just to confirm that this bug is still present in 1.10.1. Any chance you guys could try rebuilding binuitils per @kromancer 's link before you do the next version, to see if that solves the problem?

[juangburgos]

Similar issue here

Error   CMake Error:
  Running

   'C:/Program Files/Microsoft Visual Studio/2022/Community/Common7/IDE/CommonExtensions/Microsoft/CMake/Ninja/ninja.exe' '-C' 'C:/Users/admin/Desktop/REPOS/libuv.git/out/build/x64-Debug' '-t' 'restat' 'build.ninja'

  failed with:

   ninja: error: failed recompaction: Permission denied   C:\Users\admin\Desktop\REPOS\libuv.git\   ninja   

[ocroquette]

Similar issue here, using Ninja 1.10.2:

CMake Error: Running '.../ninja.exe' '-C' '... '-t' 'recompact'

  failed with:

   ninja: error: failed recompaction: Permission denied

[Ashafix]

Bit late to the party but in my case directory names with spaces caused the issue.
I've created a symlink
mklink /D "c:\ProgramFiles\NVIDIA_GPU_Computing_Toolkit" "c:\Program Files\NVIDIA GPU Computing Toolkit"
and the the environment variables CUDA_HOME and CUDA_PATH to the new directory
set CUDA_PATH="c:\ProgramFiles\NVIDIA_GPU_Computing_Toolkit\CUDA\v11.8"
set CUDA_HOME=%CUDA_PATH%

[Frisk4973]

help
C:\aseprite\build>C:\users\rgb_m\ninja-win\ninja.exe -f C:\aseprite
ninja: error: loading 'C:\aseprite': Access is denied.
if you're wondering why i kept the username in the path when making this comment, it's because it's not my actual name (obviously)

[clshortfuse]

I had to delete my build folder. No idea why, but fixed my issue.