Bug in identifier detection

[Sekky61]

Zig Version

0.14.0-dev.1573+4d81e8ee9

ZLS Version

0.14.0-dev.160+cf9b49a

Client / Code Editor / Extensions

NVIM v0.10.1 and lspconfig

Steps to Reproduce and Observed Behavior

const std = @import("std");
const print = std.debug.print;

const Enum = enum {
    source,
    @"source.fixAll",
};

pub fn main() void {
    const @"identifier with spaces in it" = Enum.source;

    switch (@"identifier with spaces in it") {
        .source => print("abc\n", .{}),
        .@"source.fixAll" => print("abcd--\n", .{}),
    }
}

Reproduction:

1. Hover over the @"identifier with spaces in it" and perform Go to definition
2. Hover over the word source in .@"source.fixAll" and perform Go to definition
3. Hover over the word fixAll in .@"source.fixAll" and perform Go to definition

Notice that:

• The go to definition does not work for the variable in the switch statement.
• The go to definition on the .@"source.fixAll" jumps to the source definition and only works before the dot.

Expected Behavior

I expected the cursor to go to the definition of the variable/enum variant.

Relevant log output

No response

[Sekky61]

As for the fix, the problem seems to be in the identifierLocFromIndex function.
It does not work with identifiers with special characters (the @"" syntax).

I propose that identifierLocFromIndex should handle the @"" syntax.
If you agree, I would like to take a look at this issue

[llogick]

I do agree that it is an issue, see

• No inlay hints/hover/completions for @"i d e n t" syntax #2023

Spoilers ahead

Apparently one can get pretty close with

diff --git a/src/analysis.zig b/src/analysis.zig
index edf0b8c5..aa144af1 100644
--- a/src/analysis.zig
+++ b/src/analysis.zig
@@ -602,16 +602,26 @@ pub fn identifierLocFromIndex(tree: Ast, source_index: usize) ?offsets.Loc {
     std.debug.assert(source_index <= tree.source.len);
 
     var start = source_index;
-    while (start > 0 and isSymbolChar(tree.source[start - 1])) {
+    while (start > 0 and (isSymbolChar(tree.source[start - 1]) or switch (tree.source[start - 1]) {
+        ' ', '.' => true,
+        else => false,
+    })) {
         start -= 1;
     }
 
     var end = source_index;
-    while (end < tree.source.len and isSymbolChar(tree.source[end])) {
+    while (end < tree.source.len and (isSymbolChar(tree.source[end]) or switch (tree.source[end]) {
+        ' ', '.' => true,
+        else => false,
+    })) {
         end += 1;
     }
 
     if (start == end) return null;
+
+    while (start < end and std.mem.indexOfScalar(u8, " .", tree.source[start]) != null) : (start += 1) {}
+    while (end > start and std.mem.indexOfScalar(u8, " .", tree.source[end - 1]) != null) : (end -= 1) {}
+
     return .{ .start = start, .end = end };
 }

and that'd cover ~99% of the use cases..

But,

const std = @import("std");
const print = std.debug.print;

const Enum = enum {
    source,
    @"source 😃 fixAll",
};

pub fn main() void {
    const @"identifier with 😃 in it" = Enum.@"source 😃 fixAll";
    const argh = @"identifier with 😃 in it";
    _ = argh;

    switch (@"identifier with 😃 in it") {
        .source => print("abc\n", .{}),
        .@"source 😃 fixAll" => print("abcd--\n", .{}),
    }
}

For a tokens based approach feel free to peek at the commits in
https://github.com/llogick/zls/tree/llogick/at-ident-syntax-fixes
https://github.com/llogick/zls/tree/dev

Cheers

[Sekky61]

Oh, I read that issue before but I didn't remember it when I encountered it today. I will take a look at it

[Techatrix]

duplicate of #2023