Running go test for gopass modifies global config and also removes mounts

[AnomalRoil]

Summary

If you have a mounted substore, e.g. "test":

gopass (/home/anomalroil/.local/share/gopass/stores/root)
└── test (/home/anomalroil/.local/share/gopass/stores/test)

and you run go test ./... somehow your config is changed and you'll loose your mount and will have to mount it again.

Steps To Reproduce

$ gopass init --store test
$ gopass mounts
$ git clone https://github.com/gopasspw/gopass && cd gopass
$ go test ./... -count=1
$ gopass mounts

Expected behavior

Running the tests should remain contained to some kind of /tmp files and not affect the global developer config for gopass.

Environment

• OS: Ubuntu 22.04
• OS version: 6.5.6-76060506-generic #202310061235~1697396945~22.04~9283e32 SMP PREEMPT_DYNAMIC Sun O x86_64 x86_64 x86_64 GNU/Linux
• gopass Version: master tip
• Installation method: from source

[dominikschulz]

Running under strace shows that go test ./... writes to the global gopass config a few times ☹️

$ GOPASS_HOMEDIR=/tmp/gp1
$ strace -f -o /tmp/gopass.strace go test ./... -count=1
$ grep "/tmp/gp1/.config/gopass/config" /tmp/gopass.strace | grep O_WRONLY
256051 openat(AT_FDCWD, "/tmp/gp1/.config/gopass/config", O_WRONLY|O_CREAT|O_TRUNC|O_CLOEXEC, 0600 <unfinished ...>
256049 openat(AT_FDCWD, "/tmp/gp1/.config/gopass/config", O_WRONLY|O_CREAT|O_TRUNC|O_CLOEXEC, 0600 <unfinished ...>
256060 openat(AT_FDCWD, "/tmp/gp1/.config/gopass/config", O_WRONLY|O_CREAT|O_TRUNC|O_CLOEXEC, 0600 <unfinished ...>
256049 openat(AT_FDCWD, "/tmp/gp1/.config/gopass/config", O_WRONLY|O_CREAT|O_TRUNC|O_CLOEXEC, 0600 <unfinished ...>
256435 openat(AT_FDCWD, "/tmp/gp1/.config/gopass/config", O_WRONLY|O_CREAT|O_TRUNC|O_CLOEXEC, 0600 <unfinished ...>
256049 openat(AT_FDCWD, "/tmp/gp1/.config/gopass/config", O_WRONLY|O_CREAT|O_TRUNC|O_CLOEXEC, 0600 <unfinished ...>

Now we need to figure out where this happens.

Edit: Actually this overwrites the entire config (O_CREATE and O_TRUNC).

[dominikschulz]

The plot thickens:

$ grep "writing config to /tmp/gp1/.config/gopass/config" /tmp/gp.log 
2023/11/25 18:20:57.225061 gitconfig/config.go:258	gitconfig.(*Config).flushRaw	writing config to /tmp/gp1/.config/gopass/config: 
2023/11/25 18:20:57.226834 gitconfig/config.go:258	gitconfig.(*Config).flushRaw	writing config to /tmp/gp1/.config/gopass/config: 
2023/11/25 18:20:57.233848 gitconfig/config.go:258	gitconfig.(*Config).flushRaw	writing config to /tmp/gp1/.config/gopass/config: 
2023/11/25 18:20:57.234426 gitconfig/config.go:258	gitconfig.(*Config).flushRaw	writing config to /tmp/gp1/.config/gopass/config: 
2023/11/25 18:20:57.278175 gitconfig/config.go:258	gitconfig.(*Config).flushRaw	writing config to /tmp/gp1/.config/gopass/config: 
2023/11/25 18:20:57.307876 gitconfig/config.go:258	gitconfig.(*Config).flushRaw	writing config to /tmp/gp1/.config/gopass/config:

[dominikschulz]

I think I got it (mostly). Not exactly sure what the root cause is, but the trigger might be related to some of your recent changes @AnomalRoil .

We should probably add ctx = config.NewNoWrites().WithConfig(ctx) to every (?) unit test.
Will work on a PR tomorrow.