Start treating function calls as initializing expressions instead of as
value expressions.
This required adding support for expression categories. Value bindings
and temporary materialization conversions are created where necessary to
transition between expression categories. For a function call with a
return slot, we speculatively create a materialized temporary before the
call and either commit to it or replace it with something else later,
once we see how the function call expression is actually used.
This change follows the direction suggested in #3133 for initializing
expressions: depending on the return type of a function, the return
value will either be initialized in-place or returned directly. This is
visible in the semantics IR, which is a little unfortunate but is
probably necessary as this is part of the semantics of the program.
---------
Co-authored-by: Chandler Carruth <chandlerc@gmail.com>
Previously, IR generated for the computation of the return type ended up
in the package's node block. Put it with the IR for the parameters for
consistency and cleanliness.
This removes BindName, putting name information directly on VarStorage.
As a side-effect of updating semantics_ir_test for this change, I also
noted that function bodies were being generated as invalid YAML so am
fixing that (just `{}` to `[]` bracketing, otherwise the test wouldn't
work anymore).
Because names are now available, I've updated lowering to use them for
vars.
In the SemIR formatter, the name is now repeated because it's a
parameter to VarStorage. I believe this is just default behavior, and
we'd have to special-case VarStorage to remove it because it's automatic
argument printing in action. On the balance, it felt like letting it
print was reasonable.
I've noted in places that the name on VarStorage is expected to be
optional, but am not adding support because I'd have no way of testing
it at present.
This also tries to restructure the command line interface to the
toolchain a bit to make it start operating more like a compiler that
could be integrated into a build system rather than primarily as
a testing tool.
1) This switches form a `dump` subcommand to a `compile` subcommand
which has "dump" actions that can be enabled within it.
2) A distinct set of compile _phases_ that match the toolchain
structure:
- `lex` to run the lexer
- `parse` to run the parser
- `check` to fully check that the code is valid
- `lower` to lower to LLVM's IR
- `codegen` to generate executable code
3) The codegen phase has two output formats: textual assembly and
a binary object. These outputs can be configured, with a default for
an object when writing to a file and more firm default for textual
assembly when writing to stdout.
4) Select and expose the use of the LLVM host detection to compute
a default code generation target in the driver so that the command
line interface can reflect this. For example, the `help` output will
include the default target.
5) The `//toolchain/codegen` library APIs have been restructured a bit
to make the code flow a bit more naturally when implementing the new
command line structure. No real changes to the logic though.
There are also some minor tweaks to the command line interface based on
trying to use the shortest names for things that still seem likely to be
learnable for users:
- Switched `target-triple` to just `target`: the "triple" component to
this name is historical and can be confusing. For example, almost all
"triple" strings have more than three components today.
- Switched to just `--output` as now the fact that it is a file can be
configured in the documentation -- it will render as `--output=FILE`.
This also adds support for two custom output filename modes. First, when
no output is specified, we now compute one in the conventional way for
compilers by removing the file extension of the input file and replacing
it with `.o` for an object file output or `.s` for an assembly file
output. This matches the behavior of Clang and GCC for example.
Second, output to stdout is enabled with the special output file name of
`-` since it is no longer the default. This also follows the convention
of most compilers and many other command line tools to use `-` as a file
name to signify using standard in/out pipes.
There are still some rough edges here that I suspect could be improved,
but this seems like a good start of switching over to a complete
argument parser.
---------
Co-authored-by: Jon Ross-Perkins <jperkins@google.com>
Co-authored-by: Lucile Rose Nihlen <luci.the.rose@gmail.com>
Co-authored-by: Richard Smith <richard@metafoo.co.uk>
When producing textual semantics IR, map each branch instruction back to
the construct that produced it and use that to determine a name for the
corresponding block label.
Don't elide the StructType node if we've already created an equivalent
type. Its spelling and location may be interesting to diagnostics,
tooling, debug information, etc.
Add a textual IR format to the toolchain.
The exact details of the format are somewhat arbitrary right now, and I
expect them to change as we refine the semantics IR model, but at the
moment they're somewhat directly following the current structure of the
IR.
Semantics tests currently test both the "raw" format, which shows the
details of the representation, and the textual format, which is somewhat
higher level. We may want to revisit that decision once the textual
format is a bit more stable, and test only the textual format in most of
these tests, but for now it seems prudent to keep both sets of tests.
---------
Co-authored-by: Chandler Carruth <chandlerc@gmail.com>
- Only allow assignment at the top level in an expression statement.
- Allow both negation and complement as subexpressions of both bitwise
and numeric operators.
- Remove parsing support for postincrement and postdecrement.
- Add parsing support for `as` operator.
- Use the same ambient precedence for types and non-type expressions.
---------
Co-authored-by: Chandler Carruth <chandlerc@gmail.com>
Refactor type canonicalization so that we can reuse the same code for
building a `T*` expression and for forming the type of an `&x`
expression.
Add basic computation of expression category in order to check that we
only take the address of durable reference expressions. This is
currently computed on demand rather than being tracked as part of the
semantics node, but in most cases can be determined by looking at only a
single expression, so caching it in the node doesn't seem worthwhile
yet. This decision should be revisited if we start doing more complex
category calculations.
Also add trivial lowering support, but it doesn't work properly yet
because lowering doesn't yet take the expression category into account.
Allow interleaving of STDOUT and STDERR check lines. Put STDOUT lines
after the line they're attached to, and STDERR lines before. If no
STDOUT check line is attached to any line, then put them all at the end
of the file instead.
This is intended to better handle the case where stdout contains
unreplaced mentions of line numbers, and also reflects that stdout is
typically a consequence of the test rather than commentary on it, so
placing it after the test seems likely to read better.
Semantic handling for use of `T*` and `const T` as types.
There's no way to form values of these types yet, and no conversions for
them are supported.
Factor out the common code to canonicalize types using a folding set,
and switch to using the same folding set for all kinds of type by adding
the kind as part of the folding set key.
Improve type printing to not include the `as type` portion when the type
is printed in a context within another type where a conversion to `type`
is implied, as in `{}*` and pre-existing cases like `({}, {}) as type`
(which we used to print as `({} as type, {} as type}) as type`.
I've migrated the toolchain autoupdate scripts here, I just need a
little more time to do the explorer side (which I need to check
performance, that may require multithreading as we do in the current
script). However, this felt substantial enough to share and it means I
can handle autoupdate in more of the toolchain, including preparatory
work for autoupdate on multi-file tests.
Once explorer is done I'll remove the old script.
For `Assign` and `ReturnExpression`, this field wasn't used for
anything. For `StructTypeField`, we stored the type of the field here,
and now store it as an argument of the node instead.
---------
Co-authored-by: Jon Ross-Perkins <jperkins@google.com>
This is sufficient to convert remaining toolchain tests to file_test. %s
and %t are currently used. While #2978 might use %T with lit, I think
that's resolving a lit-specific issue that's not necessary when
TEST_TMPDIR is readily available for any necessary operations (i.e.,
setting the working directory). With ARGS, I think it's feasible to
switch to file_test without %T.
Also adds documentation to file_test_base.h, which was starting to feel
like a significant gap.
This is not yet handled by autoupdate, but I'm eyeing that next.
Also switch from modeling type values as `i1` to modeling them as an
empty struct. We don't need any runtime representation for types, as
there are no runtime operations on them, so an empty struct seems like a
good representation.
This also makes the filename a reference to the buffer since the line
seems better to have as a reference (versus copying a string per error).
Most tests now have different line deltas due to the extra output, but
the actual errors should overall stay the same.
Some of the error locations look like they could be improved, but this
change is only making it clear where they were before.
Handles tuples (including nested tuples) in the semantic phase of the
tool chain. Does not handle tuple element access yet.
---------
Co-authored-by: Farzana Ahmed Siddique <fasiddique@google.com>
Co-authored-by: Jon Ross-Perkins <jperkins@google.com>
This started with cleaning up the remaining Name/expression type punning
in the node stack, and grew. I'm factoring out a class because we've
previously expressed the desire to factor logic out of SemanticsContext
where possible, and this seemed like a reasonable cut.
NameExpression as the first node as a QualifiedExpression allows the
qualifier handling to consider Name in one less spot, an incremental
simplification. However, the additional complexity caused by this makes
me split ApplyNameQualifier/ApplyExpressionQualifier in order to avoid
repeat checks of the parse node's kind. The logic is still largely
shared, thus a couple helper functions. I think this is all fairly well
structured in the isolated class.
I can see that we may want to avoid passing SemanticsContext as an
argument in the future if it elides a step of lookup.
This handles namespacing of functions. Parsing and semantics are changed
significantly, while lowering works without changes. Variables can't be
namespaced yet because they're dealing with patterns, and I didn't dig
through that code.
Most of the logic is done through the new name declaration stack, which
is necessary because semantics isn't quite sure where the declaration
name ends. It'd be complex for parsing to send a signal about this,
probably involving node variants and rewrites of the tree, and this
solution seems to work well. Unfortunately this means a new stack, but
that may be inevitable due to the extra information needing to be
tracked.
Note this doesn't deal with scoped lookups of non-namespace things,
which we'll need for generics. That'll probably involve pushing resolved
scopes onto a stack (or maybe just setting a singleton value?) to affect
contextual name lookup. But, I think the basics are there to make it
work when we can test the behavior.
This renames "designator expression" to "qualified expression" and adds
"qualified declaration" in order to use terminology more consistent with
C++.
Namespaces will probably need to be considered for name mangling down
the line, but this still uses the basic name.
Add validation that every code block in a function is terminated by a sequence of terminating instructions, and that terminators don't appear anywhere else in code blocks.
This required tracking whether we're in a reachable code block. That's done on the fly when we create a new code block; the new `SemanticsNodeBlockId::Unreachable` is used to represent the case where we're not actually creating a code block because we're in unreachable code.
This is just a simplification: I think these different forms are getting in the way more than they're helping, particularly as I was looking into namespace functionality. The handling in semantics can be identical, providing a more uniform behavior.
This follows the same structure as `if` expressions, except that no
result value is needed.
Semantics IR building for code blocks is also added. Rather than popping
all the node stack entries we push for statements within a code block,
change statements and declarations to not push themselves onto the
stack. We're not notionally performing work recursively within prior
statements, and we don't need their value for anything, so it seems
cleaner to not push them. This also allows statements and declarations
to determine what syntactic context they're in by peeking at the top of
the stack, though that's not used in this patch.
Add semantic analysis and semantics IR building for `if` expressions, and add the first parts of control flow handling to semantics IR. After discussion with @chandlerc, use [block arguments](https://en.wikipedia.org/wiki/Static_single-assignment_form#Block_arguments) to convey values from the two arms of the `if` to the result. For now, only a single block argument is supported, but we should revisit this as we explore more of the requirements of the Semantics IR form.
Functions can now contain multiple code blocks, so grab the entry block up-front instead of assuming the entry block will be at the top of the block stack when we reach the end of function emission.
Add trivial support for `bool` type literal, because without it we can't write testcases.
This shifts logic so that bindings are added to name lookup only after the scope is complete, removing logic around adding/removing/re-adding names in certain scopes.
This does mean that things like a function's forward declaration will need to go through an extra hoop for name conflict checks, because under this approach a function definition does conflict checking when it adds names for the body's use. But, that seems easy to address, and better than the current hoops.
This adds canonicalization of struct types based on their type fields. It obsoletes the current CanImplicitAsStruct because the type ids should now be identical when they're structurally identical; there's only a reason to implicit CanImplicitAsStruct to detect _compatible_ conversions.
The type fields themselves aren't canonicalized because it would need to be done during the first parse, and could yield name conflicts being associated with the wrong location. i.e.:
```
var x: {a: i32, a: i32};
var y: {a: i32, b: i32, a: i32};
```
This should yield two separate name conflict diagnostics pointing at the type fields for each respective line, but if struct type fields were canonicalized then both would point at the first `a: i32` field definition. This isn't expected to be an issue for types because I'm trying to print those, but we may also end up with a "first defined at" situation in some cases (still, less confusing because the type should match). Regardless, I think individual fields gets much more awkward.
This switches types to using SemanticsTypeId instead of SemanticsNodeId, and lowering pre-builds its list of types. The empty tuple type is special-cased because we don't want to emit it unless it's in-use, but as the implicit return for functions, it's frequently used. Callables use invalid to indicate the implicit return, and that seems undesirable to change due to the size increase.
This removes special-casing of empty structs, handling them as just a regular value instead of a builtin. Note the `{} as Type` is still special-cased.
In lowering, removes the test of calling a function using `{}` because it's missing the proper load/store. This setup notices that error whereas the prior worked due to said special-casing. Fixing this will need to be done as part of generally adding loads for variable uses.
This is the first step to refactoring types into a SemanticsTypeId. This only tracks what's in-use, but as a consequence starts funneling type information through in ways similar to how I'd want it to do SemanticsTypeId.
Previously, IR for arguments in calls and struct values was separated out. This merges it back in. Additionally, parameters for functions and struct types had their own IR; the block is still there, but there's a TODO to decide what to do with it.
In the LLVM IR, this has the consequence of emitting expressions that are inputs to a call or struct value within the scope of the function, which is pretty much where it should be. Importantly it happens before the call is encountered.
This change also tinkers with the int and real literal lowering. I'm pretty sure both are still wrong, but was having trouble figuring out a "better" way to do it, and this seems like it'll work for now.
Richard SmithandChandler Carruth
Jon Ross-Perkins
Farzana Ahmed SiddiqueandFarzana Ahmed Siddique