[BUG] Wrong location for ill formed definitions

[MaxSagebaum]

Title: Wrong location for ill formed definitions

Description:

A wrong definition like int a = 4; should be reported with the correct line.

Minimal reproducer (https://cpp2.godbolt.org/z/vcb5xWKxz):

func: () = {
    a : int = 5;
    int b = 4;
    c: int = 3;
}

Commands:

cppfront main.cpp2 -o main.cpp

Expected result:

main.cpp2(3,12): error: ill-formed initializer (at ' int b = 4;')
main.cpp2(1,1): error: unexpected text at end of Cpp2 code section (at 'func')
main.cpp2(1,0): error: parse failed for section starting here

Actual result and error:

main.cpp2(1,12): error: ill-formed initializer (at '{')
main.cpp2(1,1): error: unexpected text at end of Cpp2 code section (at 'func')
main.cpp2(1,0): error: parse failed for section starting here

[JohelEGP]

See also #358 (comment), #352 (comment).

[MaxSagebaum]

Thanks for the references. I could have a look at this kind of error next week and try to improve the message. Is anybody else working on this? So we do not come into conflict?

[JohelEGP]

My impression is that the logic isn't centralized,
so it'd have to be done on a case-by-case basis.

[JohelEGP]

That means there's little chance of conflict.
For example,
your reproducer fails when parsing the next statement.
#358 (comment) fails at the end of a particular expression production.
#352 (comment) failed at the end of a statement.

[MaxSagebaum]

Ok, then will try to hunt it down next week.

[MaxSagebaum]

I could track it down yesterday. The stacktrace for the parsing error is:

#0  cpp2::parser::unnamed_declaration (this=0x7fffffffc670, start=..., semicolon_required=true, captures_allowed=false, named=true, is_parameter=false, is_template_parameter=false, id=std::unique_ptr<cpp2::unqualified_id_node> = {...}, access=cpp2::accessibility::default_)
    at ./source/parse.h:5966
#1  0x0000000000431eee in cpp2::parser::declaration (this=0x7fffffffc670, semicolon_required=true, is_parameter=false, is_template_parameter=false) at ./source/parse.h:6684
#2  0x000000000042b17a in cpp2::parser::statement (this=0x7fffffffc670, semicolon_required=true, equal_sign=..., parameters_allowed=true) at ./source/parse.h:5276
#3  0x000000000042b813 in cpp2::parser::compound_statement (this=0x7fffffffc670, equal_sign=..., allow_single_unbraced_statement=false) at ./source/parse.h:5373
#4  0x000000000042b0c8 in cpp2::parser::statement (this=0x7fffffffc670, semicolon_required=true, equal_sign=..., parameters_allowed=false) at ./source/parse.h:5270
#5  0x000000000042fb08 in cpp2::parser::unnamed_declaration (this=0x7fffffffc670, start=..., semicolon_required=true, captures_allowed=false, named=true, is_parameter=false, is_template_parameter=false, id=std::unique_ptr<cpp2::unqualified_id_node> = {...}, access=cpp2::accessibility::default_)
    at ./source/parse.h:6207
#6  0x0000000000431eee in cpp2::parser::declaration (this=0x7fffffffc670, semicolon_required=true, is_parameter=false, is_template_parameter=false) at ./source/parse.h:6684
#7  0x0000000000432357 in cpp2::parser::translation_unit (this=0x7fffffffc670) at ./source/parse.h:6750
#8  0x000000000042203c in cpp2::parser::parse (this=0x7fffffffc670, tokens_=std::vector of length 24, capacity 32 = {...}, generated_tokens_=std::deque with 0 elements) at ./source/parse.h:3385
#9  0x000000000043cf14 in cpp2::cppfront::cppfront (this=0x7ffffffe65e0, filename="src/temp.cpp2") at cppfront.cpp:1157
#10 0x0000000000414a9a in main (argc=2, argv=0x7fffffffcd68) at cppfront.cpp:5926

The best position to insert it was at frame 3. Adding an error at frame 0 would have broken everything. The principle that a creation failure does not imply a parse failure makes it not possible. Adding it at frame 3 did not change anything in the regression tests.

The if prevents an error message when a more detailed message was already added to the stack and also prevents, that the same message is printed twice.