improve sysimage.mk more completely

[vtjnash]

Add back support for -debug builds and define ability to use shared
library builds more places in the file.

Fix #57675

[topolarity]

This seems to have impacted pre-compilation statements in the build.

With PR:

Collecting and executing precompile statements
└ Collect (Basic: ✓ 261) => Execute ✓ 340
Precompilation complete. Summary:
Total ───────   4.582742 seconds

Without PR:

Collecting and executing precompile statements
└ Collect (Basic: ✓ 1189) => Execute ✓ 1258
Precompilation complete. Summary:
Total ───────  38.161908 seconds

[vtjnash]

Good, that was the goal

[vtjnash]

We appear to have saved over 30 seconds there doing useless work

[topolarity]

Oh, that's great! I wasn't expecting "used shared libraries" to affect what we're pre-compiling.

What's the bit that helps?

[Keno]

I think this now requires a full basecompiler build with debug just in order to get a debug build of julia. It used to be that it could just generate the .so from the existing sysimage. I think that's a significant usability regression because that takes many, many minutes.

[vtjnash]

It used to be that it could just generate the .so from the existing sysimage.

It still could, even though this currently makes the choice towards being more reliable (since the previous design could be convinced--with a good bit of effort--to crash if used with -j), we can link the same .a file against as many different julia binaries as we want. (my normal workflow is to do cp usr/lib/julia/foo.a usr/lib/julia/foo-debug.a as part of make debug, for those rare cases where FORCE_ASSERTIONS wasn't already enough, and has been my personal workflow for years since it also solved the problem that generating the final .so is also slow and takes many, many minutes too)

I think that's a significant usability regression because that takes many, many minutes.

That is mostly an implementation regression from the bindings changes though (bootstrapping relies very heavily on some features that are currently broken due to the bindings changings), which should be improved upon soon

[topolarity]

@gbaraldi Do you know why you marked this for back-port?

I know the motivation for #57500, but not for this bit.