Previously reported as a sidenote in Issue21016.

The `--ignore-dir` option in trace.py replaces `$prefix` and `$exec_prefix` *anywhere* in the path when it really should just replace it in the start of the path and if it is followed by nothing or a path separator (that is, it is a path component). I suspect that there's always a separator though.

str.replace() can be replaced with re.sub() and \b marker.

Example:

>>> re.sub(r'\$prefix\b', '[xxx]', '$prefix $prefixpath')
'[xxx] $prefixpath'

`\b` is a bit too loose. That example should *not* have it replaced because it is not a full path component.

(Granted, any of these conflicting paths are "dumb" in general; I'm fine with just leaving this as a low priority, but if it does get modified, it should actually get fixed.)

Would you mind to add examples of patterns should be replaced and patterns which should not be replaced?

Should be replaced (using $prefix here, but equally viable for $exec_prefix). Each quoted line is a path using a raw string.

r"$prefix"
r"$prefix/foo"

and on Windows:

r"$prefix\bar"

These should *not* be replaced:

r"$prefixvar/subdir"
r"not/a/$prefix"
r"$prefix spacevar/subdir"

because here, the `$prefix` is part of another path component. This does leave the directory literally named "$prefix" in a bit of a pickle.

Like I said, it's niche, but if we're going to fix it, let's at least consider handling more cases properly.

> These should *not* be replaced:
>
> r"not/a/$prefix"
> r"$prefix spacevar/subdir"

Honestly, I don't see an issue to replace $prefix in these examples.

These examples seem artificial. The paths are not user provided: they are hardcoded paths from the sysconfig module:

    if opts.ignore_dir:
        _prefix = sysconfig.get_path("stdlib")
        _exec_prefix = sysconfig.get_path("platstdlib")

I'm not even sure if it's worth it to fix this issue, it's unclear to me how you could get r"$prefixvar/subdir" from sysconfig.get_path("stdlib").

> The paths are not user provided: they are hardcoded paths from the sysconfig module:

No, those paths are the *replacement* values, not the input. From the trace docs:

>   trace.py -c -f counts --ignore-dir '$prefix' spam.py eggs

This is the string where `$prefix` is replaced with the value retrieved from `sysconfig`.

Do you want to propose to write a PR to fix this issue?