error_with_backtrace_outputs_correctly_with_one_source fails on Windows7

[roblabla]

Since #117089 was merged, the error_with_backtrace_outputs_correctly_with_one_source std test fails with the following error on windows7:

---- error::tests::error_with_backtrace_outputs_correctly_with_one_source stdout ----
thread 'error::tests::error_with_backtrace_outputs_correctly_with_one_source' panicked at library/std/src/../../backtrace/src/dbghelp.rs:168:1:
called `Option::unwrap()` on a `None` value

After some further instrumentation, I was able to find out the source of the problem: Failed to find SymAddrIncludeInlineTrace in dbghelp. This was likely caused by rust-lang/backtrace-rs@c1464fa , which reworked the backtracing code on windows to be faster through the use of newer APIs. Unfortunately, not all of those APIs are available in Windows 7.

Version it worked on

7201240

Version with regression

608e968

@rustbot modify labels: +regression-from-stable-to-nightly -regression-untriaged

[lqd]

cc @wesleywiser #117089 is going to land in 1.75 while rust-lang/compiler-team#651 drops windows <10 support from 1.76.

Though to be clear windows 7 is not particularly well supported today, CI only tests windows 10.

[roblabla]

@lqd While this manifests as a test error, I'm pretty sure there's an actual regression here that can be observed in code, and isn't just a unit test failure. I haven't taken the time to make an actual reproducer, but I think any attempt to print a Backtrace through its Display implementation will fail. I'll try to make some code showing the regression tomorrow.

Regarding windows7 being not well-supported, I'm well-aware of that. I'm trying to improve on this by running the library/std tests on windows7 every night on some (private) infrastructure, and eventually adding a new (tier3) windows7 target once support is officially dropped in 1.76 so we can continue targeting it.

[lqd]

I agree this is likely a regression, and pinged the author: it’s likely the fact that the PR will not work on windows 7 wasn’t known then, but at the same time it seems it could be accepted if it landed today.

It’s likely the panic could be resolved though, or some of the old code brought back for the new tier3 target.

[apiraino]

Unfortunately, not all of those APIs are available in Windows 7.

@roblabla Just to be sure I understand: Do I read correctly that due to the absence of some APIs, the issue won't reproduce on Win10?

WG-prioritization assigning priority (Zulip discussion).

@rustbot label -I-prioritize +P-low

[roblabla]

That is correct, the missing APIs were added in dbghelp.dll version 6.2, which I believe comes built into Windows 8+. At least win10 has it for sure.

[roblabla]

I just realized this actually affects the 1.75 beta branch as well, this is a regression from stable to beta.

[ChrisDenton]

Once the backtrace PR is merged another PR updating the crate can be merged here with the beta-nominated label.