parsing a subtype_indication

[Ole Nielsby]

Given this lump:


entity my_entity is
port
(
my_vector : out my_vector_type(17 downto 0)
);
end my_entity;


it is clear that
my_vector_type(17 downto 0)
parses as:
subtype_indication
type_mark
type_name or subtype_name
simple_name
identifier "my_vector_type"
constraint
index_constraint
"("
discrete_range
range
simple_expression "17"
direction "downto"
simple_expression "0"
")"

What I'd like to know is, when the parser has seen:
my_vector_type(
can it, without symbol table information, infer that
the type_mark has ended and a constraint has begun?

Or can a type_name start with xxx( ?
The BNF grammar seems to allow this, allowing a
name to be a function call, but I am having a hard time
trying to find out if it is a real possibility or just lax
grammar spec.

 

[Mike Treseler]

Ole Nielsby wrote:

What I'd like to know is, when the parser has seen:
my_vector_type(
can it, without symbol table information, infer that
the type_mark has ended and a constraint has begun?

Don't know, but "my_vector_type(17 downto 0)"
is a *subtype* of some anonymous unconstrained array type
unless it is explicitly declared.

-- Mike Treseler

 

On Dec 12, 5:28 am, "Ole Nielsby" <ole.niel...@tekare-you-
spamminglogisk.dk> wrote:

What I'd like to know is, when the parser has seen:
my_vector_type(
can it, without symbol table information, infer that
the type_mark has ended and a constraint has begun?

In the general BNF grammar, no, because type_mark is a name, which
might be a slice name, which can also contain a (17 downto 0). So you
need to use a special production for type mark which allows only an
expanded name. I've found it easier to use parser combinators to
build a number of abstract parse trees, then resolve each, using
symbol table info, to a (possibly empty) concrete interpretation, then
check that the set of abstract trees yields only one valid
interpretation. You can trade off the number of parse tree branches
for the richness of your AST.

You're going to have a really hard time if you try to build a
"conventional" compiler for VHDL, employing syntax-directed
translation driven by the symtab.

- Kenn