Fix creating incorrect .elf file when building .nro

[nyanpasu64]

This project's Switch Makefile creates an appname_switch.elf file (from $OUTPUT), then attempts to convert it to appname.nro (from $BINARY). However, it builds appname.nro using libnx's switch_rules pattern, which always builds a %.nro file from an identically-named %.elf file. As a result, make would link a second .elf file named appname.elf, then delete it after converting to appname.nro. As a result, we waste more time and disk space linking twice, and gdb's sharedlibrary (share) command still looks for symbols in the missing appname.elf.

I solved this by writing a custom build rule, which builds appname.nro from appname_switch.elf. If you wanted to use the existing %.nro pattern on appname_switch.elf:

• You'd have to recursive-make to build appname_switch.nro from appname_switch.elf, then rename the file to appname.nro.
• You can't add appname_switch.nro as a prerequisite of appname.nro, since if you reran make with no changes, it would see appname_switch.nro is missing, create it again, and rename it on top of the existing appname_switch.nro.
• If you add an order-only dependency to appname_switch.nro, running make a second time still outputs appname_switch.nro, but doesn't rename it.

Discussion:

• My understanding is that .elf files are named by platform to avoid different platforms reusing the same names?
• My suggestion would be to use generic .elf names, place output files in the per-platform build folders, and place intermediate files in a subfolder under that (much like CMake build directories), but this would break current packaging scripts and produce a different directory layout from the standard Atmosphere/libnx/switch-examples makefiles.
• Why not place the .elf files in build_switch and leave appname.nro at the output root?

[vgmoose]

Thanks for this analysis, and yep your understanding is right, without being named per-platform, the wrong arch's appstore.elf would sometimes try to be linked when builds weren't cleaned back to back.

Why not place the .elf files in build_switch and leave appname.nro at the output root?

Is this an option? I haven't made big changes to this Makefile structure since it was introduced, but at the time I remember having a lot of issues fighting with those subfolders and placement of the .elf's

[nyanpasu64]

Why not place the .elf files in build_switch and leave appname.nro at the output root?

Is this an option?

I don't know if it's an option, but don't have the energy to look into it. I also do suspect that gdb may not be able to find the .elf in a subdirectory, unless you launch gdb from it. As a test I tried attaching from the parent of the Git clone, and gdb failed:

(gdb) share
warning: remote target does not support file transfer, attempting to access files from local filesystem.
warning: Could not load shared library symbols for 2 libraries, e.g. appstore_switch.elf.
Use the "info sharedlibrary" command to see the complete listing.
Do you need "set solib-search-path" or "set sysroot"?

Placing all output files in the build_switch folder would work, but may run into build system design conflicts (I'm not sure).

In any case I don't particularly want to make further changes before this is merged, unless you have an idea of what to change.