Adding LTO options

[quic-akaryaki]

No description provided.

[quic-seaswara]

Overall it LGTM with some minor nits.

Sorry for my late initial review.

I have few comments to begin with before I start doing a thorough review

• The tests added assume that they run on linux, they would need to be made UNSUPPORTED for windows

This is the most important for now, since we are at the 22.0 release point. Other comments can wait and can be done in a follow up.

• Diagnostics improvements such as the below (ld.eld --lto-obj-path=/usr/bin/ 1.o --save-temps) can be early detected before the linker ? Compare lld behavior and error out.
• Help menu options can be spelled of what kind of values they take.
• Please document the options being added with a close attention on any behavioral changes to options, I hope there is none.

Do we need to make any changes to LTO and linker script support related to these options ? Just trying to make sure.

Any changes to LTO (plugin) infrastructure ?

I will try to go over the code in detail and have comments.

[quic-akaryaki]

Overall it LGTM with some minor nits.

Sorry for my late initial review.

I have few comments to begin with before I start doing a thorough review

• The tests added assume that they run on linux, they would need to be made UNSUPPORTED for windows

I disabled one test that uses /dev/null. I don;t see other linux assumptions. The windows builder for a branch is broken, unfortunately.

This is the most important for now, since we are at the 22.0 release point. Other comments can wait and can be done in a follow up.

• Diagnostics improvements such as the below (ld.eld --lto-obj-path=/usr/bin/ 1.o --save-temps) can be early detected before the linker ? Compare lld behavior and error out.

I don't think lld does any of that.

• Help menu options can be spelled of what kind of values they take.

Options that take paths or numbers describe that in the help string. --lto-obj-path was missing the description, which I will add.

• Please document the options being added with a close attention on any behavioral changes to options, I hope there is none.

Will be done separately.

Do we need to make any changes to LTO and linker script support related to these options ? Just trying to make sure.

No.

Any changes to LTO (plugin) infrastructure ?

No.

I will try to go over the code in detail and have comments.

[quic-seaswara]

$ ld.lld --lto-obj-path=/usr/bin/ 1.o
ld.lld: error: cannot create /usr/bin/: Is a directory
LLVM ERROR: IO failure on output stream: Bad file descriptor

With your patch, we get

"UNEXPECTED LINKER BEHAVIOR"

[quic-akaryaki]

--lto-obj-path=/usr/bin/ 1.o

This happens due to report_fatal_error() in the LTO backend on the compiler side. Unfortunately, there is no graceful error handling. To fix this, we need changes in the upstream compiler code. lld does not suffer from this problem because it provides memory streams to the compiler and then saves the content in files later. eld's approach seems overall better to me as additional copy is avoided. I don't see a good fix here, other than creating a dummy file before which seems a little gross to me.

TMP=/usr/bin ld.eld 1.o may be similar (but note exactly the same).