Case broken by this PR, profile with

[replace_requires]
zlib/1.3: zlib/1.3#rrev1

We can try to be smart and think about all the other potential corner cases that would be breaking, and keep increasing the regex complexity to cover those, but there will always be more that we cannot simply foresee. Take into account that profiles also support jinja syntax, that in turn can contain practically any python code, so regex based replacer there will always be risky.

Yes, you are totally right!
I've modified the code to ignore comment function of # if it is not at least separated by a blank character from the end of the line.
All tests are passing now.

We may see if there is any other possible case where we could be breaking something.
I can't think of any at the moment.

In terms of breaking changes in the jinja python injected code... I think there can't be any issue.
E.g. In this template...

[settings]
os = {{ {"Darwin": "Macos"}.get(platform.system(), platform.system()) }}

there can't be any # inside (except inside a string -> which are correctly handled by the regex)
And there is also no possibility to put a \ at the end because a jinja template must always be surrounded by
{{ ... }} or {% ... %}

Also, and fortunately, jinja supports natively multiline blocks without needing to add \ at the end of each line so that is great!

Note, that if you want to have a beautiful formatted dict, you can probably define such a dict in jinja, then render it as text in the extra_variables field.

Closing as risk is too high for little value.