digitalmars.D - Re: problem with declaration grammar?
- Jerry Quinn <jlquinn optonline.net> Feb 20 2009
- Ellery Newcomer <ellery-newcomer utulsa.edu> Feb 20 2009
Ellery Newcomer Wrote:jerry quinn wrote:Ellery Newcomer Wrote:Maybe I'm missing something. The grammar shown in http://www.digitalmars.com/d/2.0/declaration.html has the following rules: BasicType2: * [ ] [ Expression ] [ Expression .. Expression ] [ Type ] delegate Parameters FunctionAttributesopt function Parameters FunctionAttributesopt Declarator: BasicType2 Declarator DeclaratorSuffixesopt BasicType2 Identifier DeclaratorSuffixesopt With this definition, I don't see how you can get Declarator->Identifier. Jerry
match what it does above multiple times.
As I'm looking at this further, there seems to be more issues. In particular, I don't think the grammar can parse: int (*x)(char); as specified. Doing so gives (my best attempt) Decl -> BasicType Declarators ; BasicType -> int Declarators -> DeclaratorInitializer DeclaratorInitializer -> Declarator Declarator -> BasicType2 Identifier DeclaratorSuffixes BasicType2 -> NULL (assuming that the grammar should be revised like this) Identifier -> BAD PARSE
yeah, if you haven't figured out by now, the grammar is a bunch of hooey. I spent like a month building an ANTLR grammar based on the above, and then realized much of it was garbage. Then I spent two months going through the source code and rewriting most of the rules. Just got done with it a week or two ago :) That was all version 1, but it looks the same, so if memory serves the above rules should look something like this: BasicType2_x: * [ ] [ Expression ] [ Expression .. Expression ] [ Type ] delegate Parameters FunctionAttributesopt function Paramters FunctionAttributesopt BasicType2: BasicType2_x BasicType2 BasicType2_x epsilon Declarator: BasicType2 Identifier DeclaratorSuffixesopt BasicType2 ( Declarator ) DeclaratorSuffixesopt Apologies for any BNF misuse
Cool. Do you feel like posting the whole thing somewhere? As an aside comment, it might be better from a grammar perspective to make usage of BasicType2 optional, rather than have the epsilon in the BasicType2 rule itself. Then every rule would consume at least 1 token, and _opt is the only expansion shortcut needed. In the form show, you can simplify the BasicType2 rule to BasicType2: BasicType2 BasicType2_x epsilon
Feb 20 2009
Jerry Quinn wrote:Ellery Newcomer Wrote:jerry quinn wrote:Ellery Newcomer Wrote:Maybe I'm missing something. The grammar shown in http://www.digitalmars.com/d/2.0/declaration.html has the following rules: BasicType2: * [ ] [ Expression ] [ Expression .. Expression ] [ Type ] delegate Parameters FunctionAttributesopt function Parameters FunctionAttributesopt Declarator: BasicType2 Declarator DeclaratorSuffixesopt BasicType2 Identifier DeclaratorSuffixesopt With this definition, I don't see how you can get Declarator->Identifier. Jerry
match what it does above multiple times.
int (*x)(char); as specified. Doing so gives (my best attempt) Decl -> BasicType Declarators ; BasicType -> int Declarators -> DeclaratorInitializer DeclaratorInitializer -> Declarator Declarator -> BasicType2 Identifier DeclaratorSuffixes BasicType2 -> NULL (assuming that the grammar should be revised like this) Identifier -> BAD PARSE
I spent like a month building an ANTLR grammar based on the above, and then realized much of it was garbage. Then I spent two months going through the source code and rewriting most of the rules. Just got done with it a week or two ago :) That was all version 1, but it looks the same, so if memory serves the above rules should look something like this: BasicType2_x: * [ ] [ Expression ] [ Expression .. Expression ] [ Type ] delegate Parameters FunctionAttributesopt function Paramters FunctionAttributesopt BasicType2: BasicType2_x BasicType2 BasicType2_x epsilon Declarator: BasicType2 Identifier DeclaratorSuffixesopt BasicType2 ( Declarator ) DeclaratorSuffixesopt Apologies for any BNF misuse
Cool. Do you feel like posting the whole thing somewhere?
knock yourself out http://www.personal.utulsa.edu/~ellery-newcomer/antlrd.g D version 1, ANTLR version 2, note it isn't really complete yet; I haven't gotten around to AST construction. Oh, and don't look at the lexer part. It's ugly as sin.As an aside comment, it might be better from a grammar perspective to make usage of BasicType2 optional, rather than have the epsilon in the BasicType2 rule itself. Then every rule would consume at least 1 token, and _opt is the only expansion shortcut needed. In the form show, you can simplify the BasicType2 rule to BasicType2: BasicType2 BasicType2_x epsilon
doh. I knew that.
Feb 20 2009