This upgrades LLVM in order to make progress on #10708, and it's also been awhile since we last upgraded!
The contentious point of this upgrade is that all JIT support has been removed because LLVM is changing it and we're not keeping up with it.
The main one removed is rust_upcall_reset_stack_limit (continuation of #10156),
and this also removes the upcall_trace function. The was hidden behind a
`-Z trace` flag, but if you attempt to use this now you'll get a linker error
because there is no implementation of the 'upcall_trace' function. Due to this
no longer working, I decided to remove it entirely from the compiler (I'm also a
little unsure on what it did in the first place).
Make trait lifetime parameters early bound in static fn type. Reasoning for this change is (hopefully) explained well enough in the comment, so I'll not duplicate it here. Fixes#10391.
r? @pnkfelix
I've renamed `MutableVector::mut_split(at)` to `MutableVector::mut_split_at(at)` to be coherent with ImmutableVector. As specified in the commit log, The `size_hint` method is not optimal because of #9629.
* Don't flag any address_insignificant statics as reachable because the whole
point of the address_insignificant optimization is that the static is not
reachable. Additionally, there's no need for it to be reachable because LLVM
optimizes it away.
* Be sure to not leak external node ids into our reachable set, this can
spuriously cause local items to be considered reachable if the node ids just
happen to line up
**Note**: I only tested on top of my #10670 PR, size reductions come from both change sets.
With this, [more enums are shrinked](https://gist.github.com/eddyb/08fef0dfc6ff54e890bc), the most significant one being `ast_node`, from 104 bytes (master) to 96 (#10670) and now to 32 bytes.
My own testcase requires **200MB** less when compiling (not including the other **200MB** gained in #10670), and rustc-stage2 is down by about **130MB**.
I believe there is more to gain by fiddling with the enums' layouts.
In this series of commits, I've implemented static linking for rust. The scheme I implemented was the same as my [mailing list post](https://mail.mozilla.org/pipermail/rust-dev/2013-November/006686.html).
The commits have more details to the nitty gritty of what went on. I've rebased this on top of my native mutex pull request (#10479), but I imagine that it will land before this lands, I just wanted to pre-emptively get all the rebase conflicts out of the way (becuase this is reorganizing building librustrt as well).
Some contentious points I want to make sure are all good:
* I've added more "compiler chooses a default" behavior than I would like, I want to make sure that this is all very clearly outlined in the code, and if not I would like to remove behavior or make it clearer.
* I want to make sure that the new "fancy suite" tests are ok (using make/python instead of another rust crate)
If we do indeed pursue this, I would be more than willing to write up a document describing how linking in rust works. I believe that this behavior should be very understandable, and the compiler should never hinder someone just because linking is a little fuzzy.
In #10422, I didn't actually test to make sure that the '-Z gen-crate-map'
option was usable before I implemented it. The crate map was indeed generated
when '-Z gen-crate-map' was specified, but the I/O factory slot was empty
because of an extra check in trans about filling in that location.
This commit both fixes that location, and checks in a "fancy test" which does
lots of fun stuff. The test will use the rustc library to compile a rust crate,
and then compile a C program to link against that crate and run the C program.
To my knowledge this is the first test of its kind, so it's a little ad-hoc, but
it seems to get the job done. We could perhaps generalize running tests like
this, but for now I think it's fine to have this sort of functionality tucked
away in a test.
This commit implements the support necessary for generating both intermediate
and result static rust libraries. This is an implementation of my thoughts in
https://mail.mozilla.org/pipermail/rust-dev/2013-November/006686.html.
When compiling a library, we still retain the "lib" option, although now there
are "rlib", "staticlib", and "dylib" as options for crate_type (and these are
stackable). The idea of "lib" is to generate the "compiler default" instead of
having too choose (although all are interchangeable). For now I have left the
"complier default" to be a dynamic library for size reasons.
Of the rust libraries, lib{std,extra,rustuv} will bootstrap with an
rlib/dylib pair, but lib{rustc,syntax,rustdoc,rustpkg} will only be built as a
dynamic object. I chose this for size reasons, but also because you're probably
not going to be embedding the rustc compiler anywhere any time soon.
Other than the options outlined above, there are a few defaults/preferences that
are now opinionated in the compiler:
* If both a .dylib and .rlib are found for a rust library, the compiler will
prefer the .rlib variant. This is overridable via the -Z prefer-dynamic option
* If generating a "lib", the compiler will generate a dynamic library. This is
overridable by explicitly saying what flavor you'd like (rlib, staticlib,
dylib).
* If no options are passed to the command line, and no crate_type is found in
the destination crate, then an executable is generated
With this change, you can successfully build a rust program with 0 dynamic
dependencies on rust libraries. There is still a dynamic dependency on
librustrt, but I plan on removing that in a subsequent commit.
This change includes no tests just yet. Our current testing
infrastructure/harnesses aren't very amenable to doing flavorful things with
linking, so I'm planning on adding a new mode of testing which I believe belongs
as a separate commit.
Closes#552
- Removed module reexport workaround for the integer module macros
- Removed legacy reexports of `cmp::{min, max}` in the integer module macros
- Combined a few macros in `vec` into one
- Documented a few issues
While tracking down how this function became dead, identified a spot
(@fn cannot happen) where we probably would prefer to ICE rather than
pass silently; so added fail! invocation.
Instead of forcibly always aborting compilation, allow usage of
#[warn(unknown_features)] and related lint attributes to selectively abort
compilation. By default, this lint is deny.
Instead of forcibly always aborting compilation, allow usage of
#[warn(unknown_features)] and related lint attributes to selectively abort
compilation. By default, this lint is deny.
### Rationale
There is no reason to support more than 2³² nodes or names at this moment, as compiling something that big (even without considering the quadratic space usage of some analysis passes) would take at least **64GB**.
Meanwhile, some can't (or barely can) compile rustc because it requires almost **1.5GB**.
### Potential problems
Can someone confirm this doesn't affect metadata (de)serialization? I can't tell myself, I know nothing about it.
### Results
Some structures have a size reduction of 25% to 50%: [before](https://gist.github.com/luqmana/3a82a51fa9c86d9191fa) - [after](https://gist.github.com/eddyb/5a75f8973d3d8018afd3).
Sadly, there isn't a massive change in the memory used for compiling stage2 librustc (it doesn't go over **1.4GB** as [before](http://huonw.github.io/isrustfastyet/mem/), but I can barely see the difference).
However, my own testcase (previously peaking at **1.6GB** in typeck) shows a reduction of **200**-**400MB**.
The majority of this change is modifying some of the `ast_visit` methods to return multiple values.
It's prohibitively expensive to allocate a `~[Foo]` every time a statement, declaration, item, etc is visited, especially since the vast majority will have 0 or 1 elements. I've added a `SmallVector` class that avoids allocation in the 0 and 1 element cases to take care of that.
This patchset makes warning if crate-level attribute is used at other places, obsolete attributed is used, or unknown attribute is used, since they are usually from mistakes.
Closes#3348
This is needed so that the FFI works as expected on platforms that don't
flatten aggregates the way the AMD64 ABI does, especially for `#[repr(C)]`.
This moves more of `type_of` into `trans::adt`, because the type might
or might not be an LLVM struct.
Closes#10308.
This is needed so that the FFI works as expected on platforms that don't
flatten aggregates the way the AMD64 ABI does, especially for `#[repr(C)]`.
This moves more of `type_of` into `trans::adt`, because the type might
or might not be an LLVM struct.
Issue #8763 is about improving a particular error message.
* added case & better error message for "impl trait for module"
* added compile-fail test trait-impl-for-module.rs
* updated copyright dates
* revised compile-fail test trait-or-new-type-instead
(the error message for the modified test is still unclear, but that's a different bug https://github.com/mozilla/rust/issues/8767)
* added case & better error message for "impl trait for module"
* used better way to print the module
* switched from //error-pattern to //~ ERROR
* added compile-fail test trait-impl-for-module.rs
* revised compile-fail test trait-or-new-type-instead
(the error message for the modified test is still unclear, but that's a different bug)
* added FIXME to trait-or-new-type-instead
I added a test case which does not compile today, and required changes on
privacy's side of things to get right. Additionally, this moves a good bit of
logic which did not belong in reachability into privacy.
All of reachability should solely be responsible for determining what the
reachable surface area of a crate is given the exported surface area (where the
exported surface area is that which is usable by external crates).
Privacy will now correctly figure out what's exported by deeply looking
through reexports. Previously if a module were reexported under another name,
nothing in the module would actually get exported in the executable. I also
consolidated the phases of privacy to be clearer about what's an input to what.
The privacy checking pass no longer uses the notion of an "all public" path, and
the embargo visitor is no longer an input to the checking pass.
Currently the embargo visitor is built as a saturating analysis because it's
unknown what portions of the AST are going to get re-exported.
This also cracks down on exported methods from impl blocks and trait blocks. If you implement a private trait, none of the symbols are exported, and if you have an impl for a private type none of the symbols are exported either. On the other hand, if you implement a public trait for a private type, the symbols are still exported. I'm unclear on whether this last part is correct, but librustc will fail to link unless it's in place.
I added a test case which does not compile today, and required changes on
privacy's side of things to get right. Additionally, this moves a good bit of
logic which did not belong in reachability into privacy.
All of reachability should solely be responsible for determining what the
reachable surface area of a crate is given the exported surface area (where the
exported surface area is that which is usable by external crates).
Privacy will now correctly figure out what's exported by deeply looking
through reexports. Previously if a module were reexported under another name,
nothing in the module would actually get exported in the executable. I also
consolidated the phases of privacy to be clearer about what's an input to what.
The privacy checking pass no longer uses the notion of an "all public" path, and
the embargo visitor is no longer an input to the checking pass.
Currently the embargo visitor is built as a saturating analysis because it's
unknown what portions of the AST are going to get re-exported.
This was needed to access UEFI boot services in my new Boot2Rust experiment.
I also realized that Rust functions declared as extern always use the C calling convention regardless of how they were declared, so this pull request fixes that as well.
This replaces `*` with `..` in enums, `_` with `..` in structs, and `.._` with `..` in vectors. It adds obsolete syntax warnings for the old forms but doesn't turn them on yet because we need a snapshot.
#5830
This PR improves the single-stepping experience for if-expression (no more jumping into the *else* branch before entering the *then* branch, no more jumping to the end of the *else* branch after finishing the *then* branch). Unfortunately I don't know of a straight-forward way of writing automated tests for this. Suggestions welcome!
If a function is marked as external, then it's likely desired for use with some
native library, so we're not really accomplishing a whole lot by internalizing
all of these symbols.
If a function is marked as external, then it's likely desired for use with some
native library, so we're not really accomplishing a whole lot by internalizing
all of these symbols.
I've started working on this issue and pushed a small commit, which adds a range check for integer literals in `middle::const_eval` (no `uint` at the moment)
At the moment, this patch is just a proof of concept, I'm not sure if there is a better function for the checks in `middle::const_eval`. This patch does not check for overflows after constant folding, eg:
let x: i8 = 99 + 99;
Bare functions are another example of a scalar but non-numeric
type (like char) that should be handled separately in casts.
This disallows expressions like `0 as extern "Rust" fn() -> int;`.
It might be advantageous to allow casts between bare functions
and raw pointers in unsafe code in the future, to pass function
pointers between Rust and C.
Closes#8728
Now the privacy pass returns enough information that other passes do not need to duplicate the visibility rules, and the missing_doc implementation is more consistent with other lint checks.
Previously, the `exported_items` set created by the privacy pass was
incomplete. Specifically, it did not include items that had been defined
at a private path but then `pub use`d at a public path. This commit
finds all crate exports during the privacy pass. Consequently, some code
in the reachable pass and in rustdoc is no longer necessary. This commit
then removes the separate `MissingDocLintVisitor` lint pass, opting to
check missing_doc lint in the same pass as the other lint checkers using
the visibility result computed by the privacy pass.
Fixes#9777.
As we start to move runtime components into the crate map, it's becoming harder
and harder to start the runtime from a C function as rust is embedded in another
application. Right now if you compile a rust crate as a dynamic library which is
then linked to another application, when using std::rt::start there are no I/O
local services, even though rustuv was linked against and requested. The reason
for this is that there is no top level crate map available specifying where to
find libuv I/O.
This option is not meant to be used regularly, but rather whenever compiling a
final library crate and linking it into another application. This lifts the
requirement that to get a crate map you must have the final destination be an
executable.
Since the removal of privacy from resolve, this flag is no longer necessary to
get the test runner working. All of the privacy checks are bypassed by a special
item attribute in the privacy visitor.
Closes#4947
These two attributes are no longer useful now that Rust has decided to leave
segmented stacks behind. It is assumed that the rust task's stack is always
large enough to make an FFI call (due to the stack being very large).
There's always the case of stack overflow, however, to consider. This does not
change the behavior of stack overflow in Rust. This is still normally triggered
by the __morestack function and aborts the whole process.
C stack overflow will continue to corrupt the stack, however (as it did before
this commit as well). The future improvement of a guard page at the end of every
rust stack is still unimplemented and is intended to be the mechanism through
which we attempt to detect C stack overflow.
Closes#8822Closes#10155
