Make it follow the standard protocol(s). Unfortunately that means
that some additional information will have to be carried in
the compilation context. It seems like a decent tradeoff -- but
it may be revisited in the future.
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
Ultimately, the goal is to completely separate the compilation from
the runtime to make it possible to have the interpreter/evaluator
be "just another" entity that can perform meaningful work when
given a parsed GP4 program.
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
There were significant overlaps in the names of data structures
between the compiler and the language that made it necessary
to litter the code with P4Lang.xxxx. This refactor removes that
requirement in most places (Parser is ambiguous wherever TreeSitter
is used -- cannot avoid that!)
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
In P4, parsers are considered types. Those parsers are instantiated.
The instantiated parsers are values. Previously, gp4 treated a parser
type and a parser value as identical. This PR makes that difference
clear _and_ sets the stage for the future.
TODO: Make the same distinction between control and action types and
values.
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
For types, make it optional to return a default value. While default
values for some types are meaningful, they are not meaningful for all
types (e.g., Action, Control, Parser, Parser state, etc.).
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
Support calling (invoking) a Control. There is still plenty more to
do here, but we are off to a good start.
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
The compilation code was written as a precursor for implementation
with macros. The updates in this commit make the switch.
There is still plenty to do:
1. Comment Walker.
2. Comment Macros.
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
Add support functions to check whether a (block of) statement(s) contains
any statements that should not appear in that block (e.g., switch in
an action).
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
When a function is called, if there is an (in)out parameter,
make sure that updated values are propogated to the calling
function.
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
The type system (and the value system) now include attributes
for each type (things like direction, const-ness).
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
Add support for directions on parameters and attributed types. The
latter is necessary to support the former, but is not limited to
the direction use case. An attributed type is a P4 type with attributes
(like its direction, whether it is read only, etc.) Other attributes
will be added in the future.
Changes necessary to support attributed types include:
1. Global instances tracked during compilation are not attributed
types and not simply types.
2. (future) Type checking will have to make sure that a types
attributes do not affect type compatibility.
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
By adding an expected type to the compilation context, it is
now possible for type checking to occur on keyset expressions
and return statements at the moment that they are being compiled.
Previously, it was necessary to tentatively compile them and then
typecheck afterward.
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
Refactor the class hierarchy for parser states so that
there is a parser state that acts more like a type and
one that acts more like a value (the latter, then, serves
as the base class for the other instantiated parser states).
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
Fields were not given default values when a struct was declared
without initializers.
Also, cleanup code in the helper function for binary operations
on struct instances.
Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
Will Hawkins