bpo-40334: Support type comments #19780

gvanrossum · 2020-04-29T00:18:22Z

This implements full support for # type: <type> comments, # type: ignore <stuff> comments, and the func_type parsing mode for ast.parse() and compile().

Closes we-like-parsers#95

https://bugs.python.org/issue40334

This does not yet support the "long form" for arguments with type comments. We also totally skip `# type: ignore` comments, and there is no support for func_type_input yet.

This involved a complete refactor of the part of the grammar used to recognise formal parameters in function definitions. I renamed many of the rules involved. The rewrite always includes the comma following a parameter in that argument, so that we can also include the type comment, if any (which follows the comma). The final parameter may not be followed by a comma, so we allow the comma to be omitted, but only if a close parenthesis follows instead (preceded by the type comment, if any). Some additional complications are needed because the code generator doesn't actually support alternatives that may be empty. PS. I didn't refactor the corresponding rules for lambdas, since they don't have type comments.

gvanrossum · 2020-04-29T00:31:16Z

Wait, I forgot to un-skip the test for long-form arguments and it segfaults. On it.

gvanrossum · 2020-04-29T04:13:25Z

Hm... when parsing something as simple as

def f(a): pass

it seems the end column number for a is two characters too far. I'll investigate tomorrow.

gvanrossum · 2020-04-29T17:02:52Z

Dang I already have conflicts. Looking...

lysnikolaou

Thanks for doing this! Apart from all the TYPE_COMMENTS work, I strongly feel that the arguments improvements made a significant difference in readability.

Grammar/python.gram

Parser/pegen/pegen.c

Parser/pegen/pegen.h

lysnikolaou · 2020-04-29T19:03:07Z

It seems there's a failure in test_type_comments. Was it there before as well?

gvanrossum · 2020-04-29T19:21:51Z

The test_type_comments failure is new, looking now.

gvanrossum

All your other suggestions will appear in the next commit.

Parser/pegen/pegen.c

Parser/pegen/pegen.h

lysnikolaou · 2020-04-29T22:59:58Z

Parser/pegen/pegen.c

@@ -25,7 +25,7 @@ _PyPegen_add_type_comment(Parser *p, arg_ty a, char *tc)
    if (tc == NULL) {
        return a;
    }
-    PyObject *tco = _PyPegen_new_type_comment(p, tc);
+    PyObject *tco = NEW_TYPE_COMMENT(tc);


Doesn't NEW_TYPE_COMMENT expect tc to be a Token *, i.e. that tc->bytes is a bytes object? I feel that NEW_TYPE_COMMENT would better be off as an inline function.

lysnikolaou · 2020-04-29T23:36:06Z

Also, all the calls to NEW_TYPE_COMMENT should be wrapped with CHECK_CALL_NULL_ALLOWED.

lysnikolaou · 2020-04-30T00:02:00Z

Parser/pegen/pegen.c

@@ -1483,13 +1597,13 @@ _PyPegen_get_values(Parser *p, asdl_seq *seq)

 /* Constructs a NameDefaultPair */
 NameDefaultPair *
-_PyPegen_name_default_pair(Parser *p, arg_ty arg, expr_ty value)
+_PyPegen_name_default_pair(Parser *p, arg_ty arg, expr_ty value, char *tc)


tc should be a Token * here.

lysnikolaou · 2020-04-30T00:04:19Z

Parser/pegen/pegen.c

+}
+
+arg_ty
+_PyPegen_add_type_comment_to_arg(Parser *p, arg_ty a, char *tc)


tc should be a Token * here.

lysnikolaou · 2020-04-30T00:10:25Z

To recap, because it might not be clear what my latest reviews are suggesting. I think that the following are needed:

Change the type of the tc argument in _PyPegen_add_type_comment_to_arg and _PyPegen_name_default_pair to Token *.
Wrap calls to NEW_TYPE_COMMENT with CHECK_CALL_NULL_ALLOWED.

I'd also make NEW_TYPE_COMMENT an inline function, but this is certainly not needed.

gvanrossum · 2020-04-30T02:03:40Z

Got it. I'm putting the CHECK_CALL_NULL_ALLOWED call inside NEW_TYPE_COMMENT -- does that make sense? I don't see that pattern used elsewhere.

lysnikolaou · 2020-04-30T11:17:07Z

Parser/pegen/pegen.h

+    if (tc == NULL) {
+        return NULL;
+    }
+    return CHECK_CALL_NULL_ALLOWED(p, _PyPegen_new_type_comment(p, PyBytes_AsString(tc->bytes)));


This will segfault if PyBytes_AsString fails. I think the best way to do it is assign it to a variable beforehand and check for NULL, but that means the the CHECK_NULL_ALLOWED call cannot be inside NEW_TYPE_COMMENT or it has to be called multiple times, once for each time a function is called that could potentially return NULL, which isn't much more beautiful than having it in the grammar IMHO. What do you think?

Suggested change

return CHECK_CALL_NULL_ALLOWED(p, _PyPegen_new_type_comment(p, PyBytes_AsString(tc->bytes)));

char *bytes = PyBytes_AsString(tc->bytes);

if (!bytes) {

return NULL;

}

return _PyPegen_new_type_comment(p, bytes);

Okay, I will try again. I really don't want to litter the grammar with CHECK_CALL_NULL_ALLOWED(NEW_TYPE_COMMENT(p, tc)) so I am changing NEW_TYPE_COMMENT to set p->error_indicator on every error exit.

lysnikolaou

One last thing and it's good to go.

lysnikolaou · 2020-04-30T16:19:48Z

Grammar/python.gram

+# type_expressions allow */** but ignore them
+type_expressions[asdl_seq*]:
+    | a=','.expression+ ',' '*' b=expression ',' '**' c=expression {
+        _PyPegen_seq_append_to_end(p, _PyPegen_seq_append_to_end(p, a, b), c) }


Suggested change

_PyPegen_seq_append_to_end(p, _PyPegen_seq_append_to_end(p, a, b), c) }

_PyPegen_seq_append_to_end(p, CHECK(_PyPegen_seq_append_to_end(p, a, b)), c) }

gvanrossum · 2020-04-30T17:55:00Z

@lysnikolaou Thank you so much for all your help on this one! It's been invaluable.

lysnikolaou

My pleasure!

lysnikolaou · 2020-05-01T00:12:44Z

Currently working on support for feature_version and I just realised that there is no assignment to tok->type_comments in _PyPegen_run_parser_from_file_pointer like the one in _PyPegen_run_parser_from_file_pointer. Is this okay or should it be fixed?

gvanrossum · 2020-05-01T02:43:01Z

Oops. Fixing in #19828

hauntsaninja · 2020-05-04T04:27:17Z

This doesn't seem to be able to handle function type comments where there aren't "normal" argument types preceding the *args and **kwargs types. I've opened #19894 with a fix, but it's my first time working with CPython's grammar, so apologies in advance for any crimes against parsers committed.

gvanrossum added 4 commits April 28, 2020 08:47

First steps towards supporting type comments

e17bb92

This does not yet support the "long form" for arguments with type comments. We also totally skip `# type: ignore` comments, and there is no support for func_type_input yet.

Support func_type_input in new parser

2e4126c

Support '# type: ignore <tag>'

7b446fb

gvanrossum requested a review from pablogsal as a code owner April 29, 2020 00:18

the-knights-who-say-ni added the CLA signed label Apr 29, 2020

bedevere-bot added the awaiting core review label Apr 29, 2020

gvanrossum marked this pull request as draft April 29, 2020 00:18

gvanrossum changed the title ~~bpo-40334: Support type comments~~ bpo-40334: Support type comments [WIP] Apr 29, 2020

gvanrossum added the skip news label Apr 29, 2020

gvanrossum changed the title ~~bpo-40334: Support type comments [WIP]~~ bpo-40334: Support type comments Apr 29, 2020

Fix segfault in long-form argument type comments

acb69b3

lysnikolaou mentioned this pull request Apr 29, 2020

Supporting feature_version=... we-like-parsers/cpython#124

Closed

Fix failing test_ast by refactoring param rules

1e8f7fd

gvanrossum marked this pull request as ready for review April 29, 2020 16:00

gvanrossum requested a review from lysnikolaou April 29, 2020 17:05

Merge remote-tracking branch 'origin/master' into type-comments

da253f3

lysnikolaou reviewed Apr 29, 2020

View reviewed changes

gvanrossum added 2 commits April 29, 2020 12:31

Fix bug in add_type_comment()

86d9f69

Suggestions from code review

c8a2888

gvanrossum commented Apr 29, 2020

View reviewed changes

Parser/pegen/pegen.c Show resolved Hide resolved

Parser/pegen/pegen.h Outdated Show resolved Hide resolved

lysnikolaou reviewed Apr 29, 2020

View reviewed changes

lysnikolaou reviewed Apr 30, 2020

View reviewed changes

Refactoring some things in response to review

fd2794b

lysnikolaou reviewed Apr 30, 2020

View reviewed changes

More thorough error checks in NEW_TYPE_COMMENT

d081578

lysnikolaou reviewed Apr 30, 2020

View reviewed changes

Add CHECK() around _PyPegen_seq_append_to_end() call

fecd8e8

lysnikolaou approved these changes Apr 30, 2020

View reviewed changes

gvanrossum merged commit c001c09 into master Apr 30, 2020

gvanrossum deleted the type-comments branch April 30, 2020 19:12

bedevere-bot removed the awaiting core review label Apr 30, 2020

gvanrossum mentioned this pull request May 1, 2020

bpo-40334: Ensure that tok->type_comments is set on every path #19828

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

bpo-40334: Support type comments #19780

bpo-40334: Support type comments #19780

gvanrossum commented Apr 29, 2020 •

edited

Loading

gvanrossum commented Apr 29, 2020

gvanrossum commented Apr 29, 2020

gvanrossum commented Apr 29, 2020

lysnikolaou left a comment

lysnikolaou commented Apr 29, 2020

gvanrossum commented Apr 29, 2020

gvanrossum left a comment

lysnikolaou Apr 29, 2020

lysnikolaou commented Apr 29, 2020

lysnikolaou Apr 30, 2020 •

edited

Loading

lysnikolaou Apr 30, 2020

lysnikolaou commented Apr 30, 2020 •

edited

Loading

gvanrossum commented Apr 30, 2020

lysnikolaou Apr 30, 2020

gvanrossum Apr 30, 2020

lysnikolaou left a comment

lysnikolaou Apr 30, 2020

gvanrossum commented Apr 30, 2020

lysnikolaou left a comment

lysnikolaou commented May 1, 2020 •

edited

Loading

gvanrossum commented May 1, 2020

hauntsaninja commented May 4, 2020

-    return CHECK_CALL_NULL_ALLOWED(p, _PyPegen_new_type_comment(p, PyBytes_AsString(tc->bytes)));
+    char *bytes = PyBytes_AsString(tc->bytes);
+    if (!bytes) {
+        return NULL;
+    }
+    return _PyPegen_new_type_comment(p, bytes);

	_PyPegen_seq_append_to_end(p, _PyPegen_seq_append_to_end(p, a, b), c) }
	_PyPegen_seq_append_to_end(p, CHECK(_PyPegen_seq_append_to_end(p, a, b)), c) }

bpo-40334: Support type comments #19780

bpo-40334: Support type comments #19780

Conversation

gvanrossum commented Apr 29, 2020 • edited Loading

gvanrossum commented Apr 29, 2020

gvanrossum commented Apr 29, 2020

gvanrossum commented Apr 29, 2020

lysnikolaou left a comment

Choose a reason for hiding this comment

lysnikolaou commented Apr 29, 2020

gvanrossum commented Apr 29, 2020

gvanrossum left a comment

Choose a reason for hiding this comment

lysnikolaou Apr 29, 2020

Choose a reason for hiding this comment

lysnikolaou commented Apr 29, 2020

lysnikolaou Apr 30, 2020 • edited Loading

Choose a reason for hiding this comment

lysnikolaou Apr 30, 2020

Choose a reason for hiding this comment

lysnikolaou commented Apr 30, 2020 • edited Loading

gvanrossum commented Apr 30, 2020

lysnikolaou Apr 30, 2020

Choose a reason for hiding this comment

gvanrossum Apr 30, 2020

Choose a reason for hiding this comment

lysnikolaou left a comment

Choose a reason for hiding this comment

lysnikolaou Apr 30, 2020

Choose a reason for hiding this comment

gvanrossum commented Apr 30, 2020

lysnikolaou left a comment

Choose a reason for hiding this comment

lysnikolaou commented May 1, 2020 • edited Loading

gvanrossum commented May 1, 2020

hauntsaninja commented May 4, 2020

gvanrossum commented Apr 29, 2020 •

edited

Loading

lysnikolaou Apr 30, 2020 •

edited

Loading

lysnikolaou commented Apr 30, 2020 •

edited

Loading

lysnikolaou commented May 1, 2020 •

edited

Loading