The assume intrinsic has a strong, negative impact on compile times, so
we're currently only using it in places where LLVM can simplify it to
nonnull metadata on a load intruction. Unfortunately a recent change
that fixed invalid assume calls introduce new assume calls for which
this simplification can not happen, leading to a massive regression in
compile times in certain cases.
Moving the assumptions from the middle of the function to the beginning
allows the simplification to happen again, bringing compile times back
to their old levels.
Fixes#25393
I've been working with some archives generated by MSVC's `lib.exe` tool lately,
and it looks like the embedded name of the members in those archives sometimes
have slahes in the name (e.g. `foo/bar/baz.obj`). Currently the compiler chokes
on these paths as it assumes that each file in the archive is only the filename
(which is what unix does).
This commit interprets the name of each file in all archives as a path and then
only uses the `file_name` portion of the path to extract the file to a separate
location and then reassemble it back into a new archive later. Note that
duplicate filenames are already handled, so this won't introduce any conflicts.
This fixes#24922 and #25017, and reduces the number of error messages that talk about "methods" when associated constants rather than methods are involved.
I will admit that I haven't thought very carefully about the error messages. My goal has been to make more of the messages technically correct in all situations, and to avoid ICEs. But in some cases we could probably talk specifically about "methods" rather than "items".
See:
https://sourceware.org/bugzilla/show_bug.cgi?id=4887#c9https://bugs.freedesktop.org/show_bug.cgi?id=65681
I just noticed this while talking to someone who was using
`os.environ['FOO'] = 'BAR'` in Python and since I'm learning Rust, I
was curious if it did anything special here. It looks like Rust has
an internal mutex, which helps for apps that are pure Rust, but it
will be an evil trap for someone later adding in native code (apps
like Servo and games will be at risk).
Java got this right by disallowing `setenv()` from the start.
I suggest Rust program authors only use `setenv()` early in main.
The code takes a prefix of the MD5 hash of the version string.
Since the hash command differs across GNU and BSD platforms, we scan for
the right one in the configure script.
Closes#25007
Also change several error messages to refer to "items" rather than
"methods", since associated items that require resolution during type
checking are not always methods.
There is no subtyping relationship between the types (or their non-freshened
variants), so they can not be merged.
Fixes#22645Fixes#24352Fixes#23825
Should fix#25235 (no test in issue).
Should fix#19976 (test is outdated).
As it is, save-analysis appears to return the span for the 'mut' in a declaration 'static mut identifier...' instead of the identifier. This minor change appears to fix the problem, by skipping the mut when it is present.
The enums chapter at the moment is ... weird. The examples aren't about enums, they're about structs, and most of the chapter talks about how enums don't support comparison operators by default (which is also true of other compound data types.) I think there was a story here once, but some coherency got lost in refactoring.
There are two preliminary patches here, one to combine the struct and tuple-struct chapters, and one to document unit-like structs, because enum syntax is easier to explain once you have those three. The final patch moves the enum chapter after the struct chapter, and rewrites most of it to talk about enums usefully (including covering matches on enums).
r? @steveklabnik
This makes the compatibility matrix render a little nicer on github, and also removes a note about windows support from 0.12 (Which is immaterial now that we're approaching 1.0)
Attempted to organize them in a way more relevant to what newbies
would be interested in hearing.
I am not satisfied by this at all, but by virtue of deleting old links alone I think it is an improvement.
r? @steveklabnik
