std: Prevent print panics when using TLS #29491
Conversation
Currently if a print happens while a thread is being torn down it may cause a panic if the LOCAL_STDOUT TLS slot has been destroyed by that point. This adds a guard to check and prints to the process stdout if that's the case (as we do for if the slot is already borrowed). Closes rust-lang#29488
|
r? @brson (rust_highfive has picked a reviewer for you, use r? to override) |
| // | ||
| // If, however, the actual I/O causes an error, we do indeed panic. | ||
| let result = match LOCAL_STDOUT.state() { | ||
| LocalKeyState::Uninitialized | |
There was a problem hiding this comment.
LOCAL_STDOUT.with would initialize data, so I think this branch should go together with the next arm.
EDIT: this is also fine, because if stdout ends up actually changed from the global default, it would also get initialised.
There was a problem hiding this comment.
Yeah this branch could actually go either way, but I figured that we may as well avoid initializing TLS if we're not gonna use it anyway (e.g. as you also mentioned it'd just go to the same place regardless)
|
LGTM. Would r+. |
|
Do we still use LOCAL_STDOUT? Can we kill it instead? |
|
@sfackler currently it's used to still capture the stdout of tests, but that's the only use that I know of. |
|
ping r? @brson |
|
@bors r+ |
|
📌 Commit 8588654 has been approved by |
Currently if a print happens while a thread is being torn down it may cause a panic if the LOCAL_STDOUT TLS slot has been destroyed by that point. This adds a guard to check and prints to the process stdout if that's the case (as we do for if the slot is already borrowed). Closes #29488
Currently if a print happens while a thread is being torn down it may cause a
panic if the LOCAL_STDOUT TLS slot has been destroyed by that point. This adds a
guard to check and prints to the process stdout if that's the case (as we do for
if the slot is already borrowed).
Closes #29488