[Issue 13491] DMD on Windows doesn't handle wildcards on the command line.

[via Digitalmars-d-bugs]

https://issues.dlang.org/show_bug.cgi?id=13491

Vladimir Panteleev <thecybershadow at gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
         Resolution|WONTFIX                     |---

--- Comment #5 from Vladimir Panteleev <thecybershadow at gmail.com> ---
OK, I am going to reopen this, because that is BULLSHIT.

(In reply to Walter Bright from comment #4)
> 1. the ordering of files presented to dmd matters - wildcards do not specify
> an order

This is not much more of a problem than with other languages. Yes, the first
filename is special. It dictates the name of the output executable. This is not
a problem! This is why we have -of. "gcc -o program *.c" is a common pattern
and there's no reason why it wouldn't apply to D.

And if you're talking about the problem where DMD puts one template
instantiation or whatever in whatever object file comes first on the command
line - that is an ugly implementation wart that has been causing nothing but
trouble to build systems since forever and needs to be fixed, not treated like
a feature!

> 2. directories under development tend to accumulate detritus in the form of
> other source files. With a lot of files in a directory it'll be difficult to
> notice if one or more extra ones are there.

"detritus in the form of other source files" does not appear out of nowhere -
maybe you have a habit of creating junk source files along with real ones, but
many programmers don't. Many build systems already enumerate all D files in a
"src" directory and compile them together. You know, like Dub, the package
manager we've accepted as D's official one. If this is acceptable for all the
people and projects listed on http://code.dlang.org/ , it should be acceptable
for DMD users, too.

> It's hard enough keeping the junk out of git directories

Bullshit. If anything, source control makes it much easier to keep junk out.
With "git status" you see any unversioned files you've created, and they stand
right out in any GUI. Maybe you find it "hard enough keeping the junk out of
git directories", but as we've seen, you keep coming up with innovative ways to
use git that are overcomplicated, outright broken, or even dangerous, or so I'm
not surprised.

> 3. D doesn't allow:
>     import std.*;
> for similar reasons.

Maybe that's because wildcard import in D works completely different than in
Java, C#, ActionScript, or all the other languages where it IS supported! A
wildcard import in those languages DOES NOT import all the source files in that
directory! It just adds that directory to the symbol search path. This works
for those languages because they allow only one top-level declaration per file,
and it must have the same file name as the file! So whenever the compiler
encounters an unknown symbol, it checks all wildcard import paths for the
presence of a file with the same name as that symbol.

> tl;dr wildcards to dmd are sloppy, error-prone, and have unpredictable
> behavior.

Walter Bright, time and time again, you have tried to keep all usability
features out of DMD that do not conform to your 80's workflows. Column numbers
in error messages; colored error output; and now this. And guess what? Time and
time again, after the feature does get in, everyone praises it once it appears
in the changelog. I am really tired of these arguments, and having the project
leader repeatedly WONTFIX-ing a common-sense usability thing is getting really
old really fast.

Further arguments:

1. On POSIX, wildcard expansion is done from the shell. So half of DMD users
already have and can use wildcard expansion, whether you want it or not.

2. Commands and build instructions written by people on POSIX are not going to
work for Windows users. Supporting wildcards on Windows will improve
cross-platform usability of DMD.

3. LDC on Windows already supports it. The current GDC release does not support
it because it was built without --enable-wirdcard - fairly sure this is an
unintentional omission because the MinGW compiler builds do support this.

4. It is a simple patch. Naturally it only needs to be done on Windows.

--