Depub #309
Depub #309
Conversation
|
If you're OK with me doing so, I suggest I rebase this against master and force push when #308 has merged before you review? |
| } | ||
| } | ||
|
|
||
| pub(crate) fn start_tracing() -> ThreadTracer { |
There was a problem hiding this comment.
I'm not sure if this is right. Why would starting tracing be a testing feature?
It's probably because your tool build with hardware tracing rustflags (and not software tracing)?
There was a problem hiding this comment.
Similarly for the other bits in this file.
There was a problem hiding this comment.
It's probably because your tool build with hardware tracing rustflags (and not software tracing)?
Yes, I used hw. I didn't realise that there'd be a difference here with sw? Would a bors try flush this out?
There was a problem hiding this comment.
I didn't realise that there'd be a difference here with sw?
My guess is that compile-time guards like this would give inaccurate results if the oracle uses only -C tracer=hw.
Ideally your oracle would include both kinds of tracing, but read on...
Would a bors try flush this out?
No, because we don't yet test software tracing with cranelift. I'm not sure it's ready yet, but @bjorn3 is the person to ask.
There was a problem hiding this comment.
Having looked at this again, I agree that applying this to swt.rs is incorrect. Fixed in 0703d94.
|
I think this is ready for squashing (and may as well sync the branch while we are here). |
|
OK, as soon as #308 merges, I'll sync and force push. |
These were found by a bodged version of depub.
This has to be handled a bit delicately, as repeatedly compiling ykrustc if there are errors sometimes fails/succeeds in ways that I don't really understand. It's possible that more depub'ing could be done, but something is better than nothing.
|
Rebased and squashed. |
|
bors r+ |
|
Build succeeded: |
2 participants
Needs #308 to be merged first. I'm raising this as useful context for softdevteam/depub#1.
This PR uses https://github.com/softdevteam/depub/ to reduce the visibility of quite a lot more items in yk. In some cases, we only slightly reduce visibility (
pub(crate)topub(super)) but there are a reasonable number where we get rid ofpubentirely. We're also able to gate a small number of items behind#[cfg(feature="testing")]to make clear when something is only needed for our (unusual) test setup. Overall, this PR has no functional change, but should make it a tiny little bit easier to understand the code base.