libstd trusts environment variables too much

[lilianmoraru]

I have not reviewed all the code(I needed specifically only these 2 functions) but at least home_dir() and temp_dir() are affected by this issue.

Let's take temp_dir as an example here.
It relies on the TMPDIR variable to be defined, otherwise it falls back to /tmp(not on Android).
My machine is on Linux and TMPDIR is not defined, which means that Rust falls back to /tmp, which is OK.
The issue is that if I do: export TMPDIR - now temp_dir will reliably return PathBuf::from("").
Same goes for export HOME -> home_dir -> Some(PathBuf::from("")).

As an example: it's not unusual for a bare-bones Linux container to have HOME undefined or just set to export HOME=.

I think that all the functions that have a fallback when an environment variable is not defined, should check if the string is not empty - if it is, then the functions should use the same fallback.

Pinging @alexcrichton

[nagisa]

It is not libstd trusting its environment too much. It is your environment being misconfigured, regardless of it being a common practice or not.

(Also, TMPDIR='' is a great way to make current working directory a TMPDIR, even across cwd changes; so I’m against changing anything here)

[lilianmoraru]

@nagisa The fallback seems to me like it is trying to avoid exactly this situation.
I want to find the user's home dir - why would the API tell me that it found it successfully but then return me an empty string? Although, it doesn't return an empty string if it detects that the variable was not set...
I think it is okay for an API like env_os to return an empty string, but definitely not home_dir.

The issue I have now is that I have to scroll every time through Rust's libstd to find out if after calling a function, I cannot rely on it to return the result, even though it returns Ok(), Some().

I hit this issue when I tried in my library to obtain the user's home_dir in a Linux container and then try to create a directory - for some reason(than found that it is because of this) I could not find the directory anywhere...

---

I think I understand from what point of view you are coming.
It seems to me that you think of it from the point of executing something in a terminal(or setting an empty string in any code).
If you set a variable to an empty string, than you should expect an empty string.
If you set in the terminal $HOME to an empty string, than querying it, you should expect an empty string.

My point of view is:
If you set a variable to an empty string, than something like env_os should definitely return an empty string.
But a call to home_dir or temp_dir to me should be a wrapper over system calls, not over terminal environment variables.
So, that call to getpwuid_r on Linux or GetUserProfileDirectoryW on Windows is actually the correct abstraction of the these API functions.

[nagisa]

Okay, so this issue is valid for at least $HOME as the documentation for home_dir says:

Returns the value of the 'HOME' environment variable if it is set and not equal to the empty string.

temp_dir has a similar claim:

On Windows, returns the value of, in order, the TMP, TEMP, USERPROFILE environment variable if any are set and not the empty string.

but only for Windows.^{we probably should unify the behaviour docs between window and unix here.}

So this is not your average “I dislike its semantics” bug, but “it does not work as documented” one and is therefore much more severe than initially seemed to me.

[lilianmoraru]

@nagisa I think that what the documentation says does not make the behavior correct.
I think the documentation states only what the code does.
Yes, it currently reads the HOME environment variable, but it does not mean that this is the way to go in a native programming language.
I would understand a scripting language(Bash as an example) to use the environment variables, but not a programming language that can make OS-specific low-level API calls.

[nagisa]

@lilianmoraru read what the documentation says again, please.

[lilianmoraru]

@nagisa The documentation implies that before returning $HOME, it makes sure that the variable is set(which it does) and that the string is not empty(which it doesn't), and if these conditions are not met, than getpwuid_r is called.

I understand that either the docs or the code is wrong here.

But, my point is that I am not sure that the environment variable should be read at all when you have access to OS syscalls, but at the same time, home_dir is under std::env...

[tbu-]

bash expands ~ to HOME if it is present (even if empty) and uses the same fallback as we do if it is not present.

I propose that we keep doing the same as the platform-specific tools.

[tbu-]

Python stdlib is doing the same as we do on non-Windows (haven't checked the Windows code path, it's in the same file). os.path.gethomedir, used int os.path.expanduser.

[Mark-Simulacrum]

AFAICT, the behavior as it stands for temp_dir and home_dir is the one we want based on a re-read of this issue so I'm going to close, but if I've misread something please leave a comment with the specific problem I missed. :)