By default use compressed debug in standalone userland debug files

[emaste]

Summary

The compiler supports CFLAGS=-gz=zlib to compress .debug sections in object files, libraries, and binaries. Enable it for standalone debug CFLAGS to reduce disk usage in objdir.

(This should also reduce the size of library/executable debug data in /usr/lib/debug but there is a bug in clang -- PR253942. Once the clang bug is fixed we'll get a bonus size reduction.)

[jhb]

I suppose this also works with GCC 6 and later as well? (I suspect it does.) If not, we'd need a COMPILER_FEATURE or the like to gate this on. For kernel builds, do we have to change the 'makeoptions' like in GENERIC or can we get by with doing it in some kern.*mk file?

[emaste]

hrmph
x86_64-unknown-freebsd13.0-gcc6: error: -gz is not supported in this configuration

[emaste]

add compressed-debug compiler feature

[emaste]

This comment has been deleted.

share/mk/bsd.compiler.mk

227 ↗

(On Diff #84922)

I'm not sure what GCC version added support; we could perhaps just stick it in the block above

[emaste]

For kernel builds, do we have to change the 'makeoptions' like in GENERIC or can we get by with doing it in some kern.*mk file?

Not sure off hand, I'll look at that as a subsequent change.

[emaste]

Add missing bsd.sys.mk change

[emaste]

Clang bug fixed by c4dbe899aa08f60b66426ecf63b1346bffe25ed8

[dim]

Yes, this would be very nice to have. But does this improve or worsen link performance? (I'm unsure if we'd need a toggle... yet more toggles)

[kevans]

In D29002#681658, @dim wrote:

Yes, this would be very nice to have. But does this improve or worsen link performance? (I'm unsure if we'd need a toggle... yet more toggles)

IMO let's not add a knob for it without a really compelling reason.

This LGTM, though.

[emaste]

In D29002#681658, @dim wrote:

Yes, this would be very nice to have. But does this improve or worsen link performance? (I'm unsure if we'd need a toggle... yet more toggles)

My guess is that it's a small pessimization if we have fast I/O. I'll see about a quick ad-hoc benchmark.

[emaste]

In D29002#681658, @dim wrote:

Yes, this would be very nice to have. But does this improve or worsen link performance? (I'm unsure if we'd need a toggle... yet more toggles)

My guess is that it's a small pessimization if we have fast I/O. I'll see about a quick ad-hoc benchmark.

[emaste]

I had a look at compiling libllvm's X86ISelLowering.o with various flags to see the effect:

Flags	Wall Clock	File Size
-gline-tables-only (std)	32.88	8861280
-g	39.62	27651272
-g -gz	39.41	15250672
-gline-tables-only -gz	34.66	6677312

[emaste]

And on Cirrus-CI, 28:46 to build 11877MB in /usr/obj without this change, 28:37 10498MB with the change
Before: https://cirrus-ci.com/task/4753310411390976
With change: https://cirrus-ci.com/task/5037492861665280