Fix #mod_use in toplevel

[dra27]

Regression from #9283. Fixes #9455.

This is an absolute show-stopper for 4.11!

[Octachron]

It sounds like it would be a good idea to have test?

[dra27]

I'm just about to push it!

[dra27]

That test definitely fails on trunk

[gasche]

Nice catch!

Some minor questions/remarks:

• there is another use of false in use_output, is that one correct?
• could we maybe use a labelled argument? (non-labelled boolean are a code smell as their call-sites are unreadable)
• I believe that wrap_mod was the right spelling, not wrap_mode, as I think "mod" refers to "module" here.

[gasche]

(But maybe wrap_mod could be renamed into wrap_in_module to avoid the confusion completely.)

[dra27]

Actually, the typo was the other way around - @diml mistyped wrap_mod which is the actual argument name. It reads to me as though Wojciech intended "wrap module" in 2012 when he added #mod_use.

I agree that a labelled argument or renaming would be better, but that code was fine for 8 years prior to being poked for a new feature - and the toplevel is getting a lot of attention elsewhere at the moment, anyway.

The use of false in #use_output is correct - otherwise it'll create an illegal module called (stdin), I think.

[gasche]

Yes, our comments on wrap_mod agree. (Although I see why it's confusing that I'm commenting on the diff from the other PR). Personally I would prefer to improve the code now that there was a bug, instead of waiting for the next bug. I will push a labeling commit.

[avsm]

I can confirm that a30e621 fixes the various topfind compilation issues I saw.

[gasche]

Approving the PR. The code looks sensible, the regression is confirmed to be fixed.

[gasche]

@dra27 I will let you decide if/when to merge given that I was the last to push a commit.

[ghost]

PR looks good to me too!

[dra27]

Thanks @diml - let's do it!