Unreachable errors' strings included in binary

[wooster0]

Zig Version

0.10.0-dev.4247+3234e8de3

Steps to Reproduce

pub fn main() void {
    x() catch unreachable;
}

fn x() !void {
    var y: u1 = undefined;
    switch (y) {
        0 => return error.SomeError,
        1 => return error.SomeOtherError,
    }
}

Expected Behavior

$ zig build-exe x.zig -O ReleaseSmall --strip
$ strings x | grep SomeError
$

No error strings are found in the binary.

Actual Behavior

$ zig build-exe x.zig -O ReleaseSmall --strip
$ strings x | grep SomeError
SomeError
$

A bunch of error strings are included in the binary, including some other strings:

$ strings x
attempt to unwrap error: {s}
OutOfMemory
InputOutput
TruncatedInput
Timeout
NoSpaceLeft
SystemResources
SomeOtherError
SomeError
ConnectionResetByPeer
Utf8ExpectedContinuation
LockViolation
WouldBlock
NotOpenForWriting
Utf8OverlongEncoding
LimitTooBig
FileTooBig
Utf8EncodesSurrogateHalf
Utf8InvalidStartByte
BrokenPipe
Utf8CodepointTooLarge
OperationNotSupported
MemoryMappingNotSupported
OperationAborted
Unexpected
AccessDenied
PermissionDenied
LockedMemoryLimitExceeded
DiskQuota
Expected closing ]
(msg truncated)
thread {} panic:
Unable to dump stack trace: debug info stripped
Panicked during a panic. Aborting.

It could very well be that I'm just reporting some issue that is very well known or a duplicate and it's still being worked on, but I'm just opening this to be sure.

stage1 does it better:

$ zig build-exe x.zig -O ReleaseSmall --strip -fstage1
$ strings x | grep SomeError
$ strings x
j	Xj
Zj"AZj
Linker: LLD 15.0.0
.rodata
.text
.bss
.comment
.shstrtab

It still includes "Linker: LLD 15.0.0" which ultimately is something that should go away as well, but it's a lot better than stage2 still. Probably if we use our own linker we can fix that. Maybe it's illegal to strip it if we use LLD.

[wooster0]

About the backend-llvm label, doesn't it affect the in-house backends and output formats too?

$ cat minimal.zig
pub fn main() void {}
$ stage3/bin/zig build-exe minimal.zig -ofmt=plan9 -OReleaseSmall
$ wc -c minimal
1959 minimal
$ hexdump -C minimal | grep -n10 featureS
107-000006a0  ff ff fa 00 00 00 00 00  00 00 00 20 00 28 f4 5f  |........... .(._|
108-000006b0  73 74 61 72 74 32 00 00  00 00 00 00 20 00 28 d4  |start2...... .(.|
109-000006c0  5f 73 74 61 72 74 00 00  00 00 00 00 20 00 2f f4  |_start...... ./.|
110-000006d0  63 61 6c 6c 4d 61 69 6e  32 00 00 00 00 00 00 20  |callMain2...... |
111-000006e0  00 6e f4 65 78 69 74 32  00 00 00 00 00 00 00 00  |.n.exit2........|
112-000006f0  01 fa 00 00 01 00 07 00  08 00 09 00 04 00 0a 00  |................|
113-00000700  00 00 00 00 00 7f ff ff  ff fa 00 00 00 00 00 00  |................|
114-00000710  00 00 20 00 8f f4 6d 61  69 6e 00 00 00 00 00 00  |.. ...main......|
115-00000720  00 00 01 fa 00 00 01 00  02 00 03 00 04 00 05 00  |................|
116-00000730  0b 00 00 00 00 00 00 7f  ff ff ff fa 00 00 00 00  |................|
117:00000740  00 00 00 00 20 00 9a f4  66 65 61 74 75 72 65 53  |.... ...featureS|
118-00000750  65 74 00 00 00 00 00 00  20 02 4e f4 65 6d 70 74  |et...... .N.empt|
119-00000760  79 5f 77 6f 72 6b 61 72  6f 75 6e 64 00 00 00 00  |y_workaround....|
120-00000770  00 00 20 03 16 f4 61 64  64 46 65 61 74 75 72 65  |.. ...addFeature|
121-00000780  00 01 02 85 42 04 bd 44  00 00 00 00 4a 9f 00 ff  |....B..D....J...|
122-00000790  ff ff b6 89 06 c0 c0 c0  c0 c0 c0 b2 46 02 c0 c0  |............F...|
123-000007a0  c0 86 42 04 c0 c0 a8                              |..B....|
124-000007a7

All the backends seem to scan all decls and keep all of them instead of keeping only the referenced one.

Or Wasm...

$ stage3/bin/zig build-exe minimal.zig -ofmt=wasm -OReleaseSmall -target wasm32-wasi -fno-LLVM
$ wasm2wat minimal.wasm
(module
  (type (;0;) (func))
  (type (;1;) (func (param i32)))
  (import "wasi_snapshot_preview1" "proc_exit" (func (;0;) (type 1)))
  (func (;1;) (type 0)
    return)
  (func (;2;) (type 0)
    call 1
    i32.const 0
    call 0)
  (func (;3;) (type 1) (param i32)
    (local i32)
    block  ;; label = @1
      local.get 0
      i32.eqz
      local.set 1
      block  ;; label = @2
        local.get 1
        br_if 0 (;@2;)
        br 1 (;@1;)
      end
      unreachable
    end
    return)
  (table (;0;) 0 funcref)
  (memory (;0;) 17)
  (global (;0;) (mut i32) (i32.const 1048576))
  (export "_start" (func 2))
  (export "memory" (memory 0))
  (data (;0;) (i32.const 1048576) "\00\05\00\00\00\00\00\00\04\00\00\00\00\00\00\00$\00\00\00\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa\aa"))

Seems like some pretty fundamental issue.

[Vexu]

All the backends seem to scan all decls and keep all of them instead of keeping only the referenced one

That is #6256

[wooster0]

Will leave this open until either there is a solution or it becomes an acknowledged and permanent flaw in the language.

Would be pretty sad.

[wooster0]

This seems to be fixed. Bloat seems to have gone down.

~$ cat x.zig
pub fn main() void {
    x() catch unreachable;
}

fn x() !void {
    var y: u1 = undefined;
    _ = &y;
    switch (y) {
        0 => return error.SomeError,
        1 => return error.SomeOtherError,
    }
}
~$ zig build-exe x.zig -O ReleaseSmall
~$ strings x | grep Error
~$

There are still other undesired strings included but that is more or less tracked by #22270