Fix _pipepager()/_tempfilepage() to work with multi-call binaries

[rhizoome]

Programs such as BusyBox, Toybox and Coreutils (also gzib, bzip etc) in multi-call mode derive their identity from the symlink. Resolving the symlink causes them to misbehave.

fixes #2943

[stefreak]

Makes sense, this looks good to me 👍

[Rowlando13]

This looks like this will work but it's not obvious that calling .resolve() will break things for BusyBox, etc. Please add a comment and a test.

[stefreak]

@Rowlando13 How can we ensure the presence of Busybox for the test?

[Rowlando13]

The issue says that it is from resolving symlinks and that multiple utilities exhibit the behavior. I would imagine one of those utilities would installed on A GitHub runner, but I don't know that they are. I was going to leave it up to the person who submitted the issue.

[rhizoome]

Would be simulating a multicall-binary in shell or as a python script good enough? That is what I planned for the test. Multicall-binaries read argv[0] and we can do that in shell or python. But of course running the whole test-suite against busybox would prevent any kind busybox based bugs in the future. I‘ll look into that too.

[Rowlando13]

I think in Python would be best. I would guess it is hard to guarantee everything that a shell gives you from inside pytest since it does a lot of funny things, but I don't know for sure. @davidism might know. I am pretty sure we don't want to have busybox as part of our test suite, since it would make things slower and I don't think any of the core contributors have experience with it.

[davidism]

It's fine without a test. Please make the comment very specific about why we use absolute and not resolve, so that we don't change it back later.

[rhizoome]

It's fine without a test. Please make the comment very specific about why we use absolute and not resolve, so that we don't change it back later.

I updated this PR, but I got curious on how difficult it is to add busybox tests. It seems quite straight forward so I added this proposal: #3101

[davidism]

Comment looks good. I'm not going to add the busybox tests, but it's a good reference, thanks for looking into it.