Add a failure checkpoint after the typechecking pass. There are still many
fatal errors in typeck, but loosening up this one makes it easier to lean on
the compiler when making changes to types.
Issue #440.
src/comp/syntax is currently just a sub-module of rustc, but it will,
in the near future, be its own crate. This includes:
- The AST data structure
- The parser
- The pretty-printer
- Visit, walk, and fold
- The syntax extension system
- Some utility stuff that should be in the stdlib*
*) Stdlib extensions currently require a snapshot before they can be
used, and the win build is very broken right now. This is temporary
and will be cleaned up when one of those problems goes away.
A lot of code was moved by this patch, mostly towards a more organized
layout. Some package paths did get longer, and I guess the new layout
will take some getting used to. Sorry about that!
Please try not to re-introduce any dependencies in syntax/ on any of
the other src/comp/ subdirs.
This way, the pretty-printer does not have to know about middle::ty.
(This is a preparation for separating the AST functionality into a
separate crate.)
An expression like:
foo(1, fail, 2)
was failing to parse, because the parser was interpreting the comma
as the start of an expression that was an argument to fail, rather
than recognizing that the fail here has no arguments
Fixed this by using can_begin_expr to determine whether the next
token after a fail token suggests that this is a nullary fail or a
unary fail.
In addition, when translating calls, check before translating each
argument that the block still isn't terminated. This has the effect
that if an argument list includes fail, the back-end won't keep trying
to generate code for successive arguments and trip the !*terminated
assertion.
The code for translating a fail (for example) would call
Unreachable(), which terminates the block; if a fail appeared as an
argument, this would cause an LLVM assertion failure. Changed
trans_call to handle this situation correctly.
Only link attributes of the meta_list type are considered when matching crate
attributes. Instead of doing nothing we can at least log that link attributes
of other types were ignored.
(Using the * operator.)
This makes tags more useful as nominal 'newtype' types, since you no
longer have to copy out their contents (or construct a cumbersome
boilerplate alt) to access them.
I could have gone with a scheme where you could dereference individual
arguments of an n-ary variant with ._0, ._1, etc, but opted not to,
since we plan to move to a system where all variants are unary (or, I
guess, nullary).
This is a preparation for tags-as-nominal-types. A tag that has only a
single variant is now represented, at run-time, as simply a tuple of
the variant's parameters, with the variant id left off.
This is important since we are going to be making functions noncopyable
soon, which means we'll be seeing a lot of boxed functions.
(*f)(...) is really just too heavyweight.
Doing the autodereferencing was a very little bit tricky since
trans_call works with an *lval* of the function whereas existing
autoderef code was not for lvals.
Modify typestate to check for unused variables and emit warnings
where relevant. This exposed a (previously harmless) bug in
collect_locals where outer functions had bit-vector entries
for init constraints for variables declared in their inner
nested functions. Fixing that required changing collect_locals to
use visit instead of walk -- probably a good thing anyway.
The parser needs to parse unconfigured items into the AST so that they can
make the round trip back through the pretty printer, but subsequent passes
shouldn't care about items not being translated. Running a fold pass after
parsing is the lowest-impact way to make this work. The performance seems
fine.
Issue #489
This represents the compilation environment, defined as AST meta_items, Used
for driving conditional compilation and will eventually replace the
environment used by the parser for the current conditional compilation scheme.
Issue #489
Tydescs are currently re-created for each compilation unit (and I
guess for structural types, they have to be, though the duplication
still bothers me). This means a destructor can not be inlined in the
drop glue for a resource type, since other crates don't have access to
the destructor body.
Destructors are now compiled as separate functions with an external
symbol that can be looked up in the crate (under the resource type's
def_id), and called from the drop glue.
Instead of linking directly to the rust crate, try to figure out the location
and name of the library from the file name, then call gcc with appropriate -L,
-l flags. This will allow dynamic linking to be more forgiving about where it
loads the library from at runtime - currently a stage3 compiler can't run
correctly from the stage0 directory. Only tested on Linux. Fingers crossed.
It would be better to either convert ast_map to use smallintmap or make
smallintmap and hashmap follow the same interface, but I don't feel up to
it just now. Closes#585.
Implement "claim" (issue #14), which is a version of "check" that
doesn't really do the check at runtime. It's an unsafe feature.
The new flag --check-claims turns claims into checks automatically --
but it's off by default, so by default, the assertion in a claim
doesn't execute at runtime.
The meta items within a crate's link attribute are used in linkage:
#[link(name = "std",
vers = "1.0",
custom = "whatever")];
Name and vers are treated specially, and everything else is hashed together
into the crate meta hash.
Issue #487
Resources are now defined like...
resource fd(int n) { close(n); }
Calling fd with an int will then produce a non-copyable value
that, when dropped, will call close on the given int.
Wrote some small test cases that use while loops and moves, to
make sure the poststate for the loop body gets propagated into the
new prestate and deinitialization gets reflected.
Along with that, rewrite the code for intersecting states. I still
find it dodgy, but I guess I'll continue trying to add more tests.
Also, I'll probably feel better about it once I start formalizing
the algorithm.
Includes assignment operations. Add regression tests for lots of less useful,
less used or unexpected combinations, as well as a selection of compile-fail
tests. Closes#500 (again!)
This will probably need more work, as moving doesn't appear to do
quite the right thing yet in general, and we should also check
somewhere that we're not, for example, moving out the content out of
an immutable field (probably moving out of fields is not okay in
general).
Non-copyability is not enforced yet, and something is still flaky with
dropping of the internal value, so don't actually use them yet. I'm
merging this in so that I don't have to keep merging against new
patches.
Modified typestate to throw away any constraints mentioning a
variable on the LHS of an assignment, recv, assign_op, or on
either side of a swap.
Some code cleanup as well.
Typeck and trans used to, by historical coincidence, use the item_obj
node id, which was used to identify the obj type by the rest of the
system, for the constructor function. This is now identified by the
ctor id stored in the tag throughout.
If you use a function expecting an alias argument in a context that
expects a function expecting a value argument, or vice versa, the
previous error message complained that the number of arguments was
wrong. Fixed the error message to be accurate.
typestate now drops constraints correctly in the post-state of
a move expression or a declaration whose op is a move. It doesn't
yet drop constraints mentioning variables that get updated.
To do this, I had to change typestate to use trit-vectors instead
of bit-vectors, because for every constraint, there are three
possible values: known-to-be-false (e.g. after x <- y, init(y) is
known-to-be-false), known-to-be-true, and unknown. Before, we
conflated known-to-be-false with unknown. But move requires them
to be treated differently. Consider:
(program a)
(a1) x = 1;
(a2) y <- x;
(a3) log x;
(program b)
(b1) x = 1;
(b2) y <- z;
(b3) log x;
With only two values, the postcondition of statement a2 for
constraint init(x) is the same as that of b2: 0. But in (a2)'s
postcondition, init(x) *must* be false, but in (b2)'s condition,
it's just whatever it was in the postcondition of the preceding statement.
