stage2: Add "tilde pointing" to errors for multi-character offenses

[greenfork]

Error refactoring in #9117 removed this style from main.zig file. This commit adds it back and in addition to all other possible compiler errors.

I added the lexeme_length to Compilation.AllErrors.Message in order to get more reliable data where it can be done. As an alternative we can use a similar approach as for Compilation.AllErrors.add for all errors since Message has both source_line and column present. But it doesn't look very good to me.

Sketchy implementation in Compilation.AllErrors.add is inspired by
https://github.com/ziglang/zig/blob/master/lib/std/zig/ast.zig#L97-L116

I tried to test all paths with the following snippets, zig run:

conststd=@import("std");
pubfnmain() void {
hello();
}

Compilation_add.zig:4:5: error: use of undeclared identifier 'hello'
hello();
~~~~~

---

conststd=@import("std");
pubfnmain() void {
h();
}

single_caret.zig:4:5: error: use of undeclared identifier 'h'
h();
^

---

conststd=@import("std");
pubfnmain() !void {
consthello=true||false;
}

addZir_with_token_present.zig:4:11: error: unused local constant
const hello = true || false;
~~~~~

---

conststd=@import("std");
pubfnmain() !void {
suspend {
suspend {}
}
// const hello = true || false;
}

addZir_with_node_present.zig:5:9: error: cannot suspend inside suspend block
suspend {}
~~~~~~~
addZir_with_node_present.zig:4:5: note: other suspend block here
suspend {
~~~~~~~

[andrewrk]

Here's a fun test case, how does this look?

test {
constx: i32=1;
consty: i32=1;
foo(x+y);
}
fnfoo(x: bool) void {
_=x;
}

Related:

test {
constx: i32=1;
consty: i32=1;
foo(x+// x is the first operandy// y is the second operand
);
}
fnfoo(x: bool) void {
_=x;
}

I have a feeling tilde pointing is going to be a bit more complicated than this implementation. We might want to enable it on a per-error-message basis.

[greenfork]

Thank you for these counter-examples. Funnily enough they produce same error which also doesn't work as expected:

start.zig:85:9: error: struct 'counter_example1.counter_example1' has no member named 'main'
root.main();
^

and if I change it to use pub fn main then these errors:

counter_example1.zig:4:11: error: expected bool, found i32
foo(x + y);
^

counter_example2.zig:4:11: error: expected bool, found i32
foo(x + // x is the first operand
^

So yes, this looks like a lot more work to do. I think I don't have enough understanding to implement it. I will try to come up with a list of questions and maybe attend a weekly stage 2 meeting.

[andrewrk]

Thank you for these counter-examples. Funnily enough they produce same error which also doesn't work as expected:

Ah the instructions for running the example would be: zig test foo.zig

[greenfork]

More on formatting:

• Rust uses ^~~~ style instead of ~~~~, examples: Multi-line errors rust-lang/rust#19870. Same for gcc and clang
• Ascii art from Rust for errors spanning multiple lines: Show multiline spans in full if short enough rust-lang/rust#37369

Examples from gcc and clang for multiline errors:

$ cat me.c
int main() {
return "a" /* comment */
+ /* comment */
1;
}
$ gcc me.c
me.c: In function ‘main’:
me.c:3:5: warning: returning ‘char *’ from a function with return type ‘int’ makes integer from pointer without a cast [-Wint-conversion]
2 | return "a" /* comment */
| ~~~~~~~~~~~~~~~~~
3 | + /* comment */
| ^~~~~~~~~~~~~~~~~~~~~~~~
4 | 1;
| ~
$ clang me.c
me.c:3:5: warning: adding 'int' to a string does not append to the string [-Wstring-plus-int]
+ /* comment */
^~~~~~~~~~~~~~~~~~~~~~~~
me.c:3:5: note: use array indexing to silence this warning
+ /* comment */
^
[
me.c:2:12: warning: incompatible pointer to integer conversion returning 'char *' from a function with result type 'int' [-Wint-conversion]
return "a" /* comment */
^~~~~~~~~~~~~~~~~
2 warnings generated.

[greenfork]

I will close this PR for now as it appears quite complex. I don't know whether there should be an issue created or there will be a more broad RFC-like issue such as "friendlier errors" so I will leave it as is.