Takes option -o into account when compiling .c files

[vbgl]

Mantis issue 6475

[yallop]

This seems to depend on the order in which the options are given. With -o first it works:

$ ls
test.c
$ ocamlc -o output.o -c test.c 
$ ls
output.o  test.c

but with -o last it doesn't:

$ ls 
test.c
$ ocamlc -c test.c -o output.o
$ ls
test.c  test.o

[vbgl]

Thanks. You are right. However this behaviour is consistent with what happens to .ml files: if the -o option comes before the input file name, it is taken into account; otherwise it isn’t.

Command-line arguments are handled in order, from left to right. When a file name is encountered, it is compiled; and then the processing of the arguments goes on. Changing this behaviour, i.e., deferring compilation after all arguments are parsed would be a much more dramatic change.

[yallop]

That's interesting. I hadn't noticed that the placement of -o mattered for .ml files as well. That's different from the behaviour of the C compilers I've used.

One useful property of the current behaviour is that it works for multiple files: you can set the output name for each file independently:

$ ocamlc -c       \
   -o a_out.o a.c \
   -o b_out.o b.c \
   -o c_out c.ml  \
   -o d_out d.ml

In contrast, gcc just says:

gcc: fatal error: cannot specify -o with -c, -S or -E with multiple files

[gasche]

could you name the new parameter with a label ~output or ~output_name to be able to distinguish both parameters better at call sites?

[whitequark]

@gasche This PR has been updated.

I have attached the patch for ocamlbuild in mantis.

[gasche]

Merged in trunk, thanks!

[damiendoligez]

There's a problem with this change. See Mantis PR#6475.

[damiendoligez]

We need to decide before 4.03 whether to revert this change.

[damiendoligez]

Change reverted, pending a more ambitious overhaul of the command-line parsing (#464).