Ralph (Mar 06 2024 at 13:47): Ralph (Mar 06 2024 at 13:47):Anyone here have experience using lldb and gdb with the wasmtime serve command? I can get lldb and gdb started, but the objective is NOT to debug wasmtime itself, but rather the running wasm and using dwarf symbols to map that to the original source code.
Ralph (Mar 06 2024 at 13:47):I can get gdb and lldb started easily enough, but because the serve command doesn't seem to open the endpoint, I can't hit it.
Ralph (Mar 06 2024 at 13:48):I do get a wonderful list of tokio processes, however, so I've got that going for me.....
Ralph (Mar 06 2024 at 13:50):I have this to follow: https://docs.wasmtime.dev/examples-rust-debugging.html, though my source code is C, so I'll need to ensure that /opt/wasi-sdk/bin/clang is building dwarf in of course.
Scott Waye (Mar 09 2024 at 02:02):NOt tried serve but for source debugging I do
c:\"Program Files"\llvm\bin\lldb target\debug\wasmtime.exe
Then
process launch --stop-at-entry -- -D debug-info c:\tmp\console\bin\Debug\net8.0\wasi-wasm\native\console.wasm
breakpoint set --file 1.c --line 5
c
do you get dwarf symbols in return? I can GET dwarf info from core modules, but seemingly I'm defied by compoents:
Ralph (Mar 10 2024 at 19:25):Core wasm:
Ralph (Mar 10 2024 at 19:25):squillace@idiopath:~/work/witlldb/hello-c$ lldb -- wasmtime -D debug-info=y foo.wasm
Traceback (most recent call last):
File "<string>", line 1, in <module>
ModuleNotFoundError: No module named 'lldb.embedded_interpreter'
(lldb) target create "wasmtime"
Current executable set to 'wasmtime' (x86_64).
(lldb) settings set -- target.run-args "-D" "debug-info=y" "foo.wasm"
(lldb) breakpoint set --name main
Breakpoint 1: where = wasmtime`main, address = 0x00000000004af180
(lldb) run
Process 181137 launched: '/home/squillace/.wasmtime/bin/wasmtime' (x86_64)
Process 181137 stopped
thread #1, name = 'wasmtime', stop reason = breakpoint 1.1
<snip>
(lldb) c
Process 181137 resuming
1 location added to breakpoint 1
Process 181137 stopped
thread #1, name = 'wasmtime', stop reason = breakpoint 1.2
frame #0: 0x00007ffff76eb60a JIT(0x7ffff732e010)`foo::main::h6f744fe0c842de61 at foo.rs:2:3
1 fn main() {
-> 2 println!("hello from everywhere!");
3 }
Ralph (Mar 10 2024 at 19:25):no problem there
Ralph (Mar 10 2024 at 19:27):but with components, I can't grab the function name. The core wasm name was "main" but the component export function is "__main_void". I tried both:
Ralph (Mar 10 2024 at 19:29):squillace@idiopath:~/work/witlldb/hello-c$ lldb -- wasmtime run -D debug-info=y comfoonent.wasm
(lldb) target create "wasmtime"
Current executable set to 'wasmtime' (x86_64).
(lldb) settings set -- target.run-args "run" "-D" "debug-info=y" "comfoonent.wasm"
(lldb) breakpoint set -name __main_void
(lldb) run
Process 182996 launched: '/home/squillace/.wasmtime/bin/wasmtime' (x86_64)
hello from everywhere!
Process 182996 exited with status = 0 (0x00000000)
(lldb) exit
squillace@idiopath:~/work/witlldb/hello-c$ lldb -- wasmtime run -D debug-info=y comfoonent.wasm
(lldb) target create "wasmtime"
Current executable set to 'wasmtime' (x86_64).
(lldb) settings set -- target.run-args "run" "-D" "debug-info=y" "comfoonent.wasm"
(lldb) breakpoint set --name main
(lldb) run
<snikp>
(lldb) c
Process 183439 resuming
hello from everywhere!
Process 183439 exited with status = 0 (0x00000000)
Ralph (Mar 10 2024 at 19:32):trying to do it by setting lines and files ALSO doesn't work, but does for core wasm:
Ralph (Mar 10 2024 at 19:32):squillace@idiopath:~/work/witlldb/hello-c$ lldb -- wasmtime run -D debug-info=y comfoonent.wasm
(lldb) settings set -- target.run-args "run" "-D" "debug-info=y" "comfoonent.wasm"
(lldb) breakpoint set -f foo.rs -l 2
Breakpoint 1: no locations (pending).
WARNING: Unable to resolve breakpoint to any actual locations.
(lldb) run
Process 187505 launched: '/home/squillace/.wasmtime/bin/wasmtime' (x86_64)
hello from everywhere!
Process 187505 exited with status = 0 (0x00000000)
(lldb) exit
squillace@idiopath:~/work/witlldb/hello-c$ lldb -- wasmtime -D debug-info=y foo.wasm
(lldb) settings set -- target.run-args "-D" "debug-info=y" "foo.wasm"
(lldb) breakpoint set -f foo.rs -l 2
Breakpoint 1: no locations (pending).
WARNING: Unable to resolve breakpoint to any actual locations.
(lldb) run
Process 187877 launched: '/home/squillace/.wasmtime/bin/wasmtime' (x86_64)
1 location added to breakpoint 1
Process 187877 stopped
Ralph (Mar 10 2024 at 19:54):And I can wire that stuff up now.... But the components elude me.
Ralph (Mar 10 2024 at 19:54):
Scott Waye (Mar 10 2024 at 23:41):I see, its compoents. I've not got there yet, but I will also be looking at it when I can get https://github.com/bytecodealliance/wasmtime/pull/8055 into a fit state.
Ralph (Mar 11 2024 at 13:11):that is a very good thing to do
Ralph (Mar 11 2024 at 13:12):so far as I can see, the current tooling strips out dwarf
Ralph (Mar 11 2024 at 13:12):if I do the core module:
Ralph (Mar 11 2024 at 13:12):squillace@idiopath:~/work/hello-wasi-http-go$ wasm-tools objdump main.wasm
types | 0xa - 0x63 | 89 bytes | 14 count
imports | 0x66 - 0x2f2 | 652 bytes | 13 count
functions | 0x2f4 - 0x342 | 78 bytes | 77 count
tables | 0x344 - 0x349 | 5 bytes | 1 count
memories | 0x34b - 0x34e | 3 bytes | 1 count
globals | 0x350 - 0x362 | 18 bytes | 3 count
exports | 0x365 - 0x473 | 270 bytes | 14 count
elements | 0x475 - 0x47e | 9 bytes | 1 count
data count | 0x480 - 0x481 | 1 bytes | 1 count
code | 0x485 - 0xa329 | 40612 bytes | 77 count
data | 0xa32c - 0xae8e | 2914 bytes | 3 count
custom "name" | 0xae96 - 0xba8e | 3064 bytes | 1 count
custom ".debug_info" | 0xba9e - 0x1fd2a | 82572 bytes | 1 count
custom ".debug_pubtypes" | 0x1fd3d - 0x21dc1 | 8324 bytes | 1 count
custom ".debug_loc" | 0x21dd0 - 0x268d7 | 19207 bytes | 1 count
custom ".debug_ranges" | 0x268e8 - 0x277c0 | 3800 bytes | 1 count
custom ".debug_aranges" | 0x277d1 - 0x27821 | 80 bytes | 1 count
custom ".debug_abbrev" | 0x27832 - 0x28457 | 3109 bytes | 1 count
custom ".debug_line" | 0x28467 - 0x2ecb6 | 26703 bytes | 1 count
custom ".debug_str" | 0x2ecc5 - 0x3e2a5 | 62944 bytes | 1 count
custom ".debug_pubnames" | 0x3e2b9 - 0x4ca6e | 59317 bytes | 1 count
custom "producers" | 0x4ca7b - 0x4cb0e | 147 bytes | 1 count
custom "target_features" | 0x4cb20 - 0x4cb5e | 62 bytes | 1 count
Ralph (Mar 11 2024 at 13:12):in short, I got the debug info
Ralph (Mar 11 2024 at 13:13):but objdump doesn't show anything in the component version of the same core module:
Ralph (Mar 11 2024 at 13:13):component imports | 0x154f - 0x1572 | 35 bytes | 1 count
module | 0x1576 - 0x1522a6 | 1379632 bytes | 1 count
------ start module 0 -------------
types | 0x1581 - 0x17cd | 588 bytes | 69 count
imports | 0x17d0 - 0x1bf7 | 1063 bytes | 24 count
functions | 0x1bfa - 0x1ed3 | 729 bytes | 727 count
tables | 0x1ed5 - 0x1eda | 5 bytes | 1 count
memories | 0x1edc - 0x1edf | 3 bytes | 1 count
globals | 0x1ee1 - 0x1ef3 | 18 bytes | 3 count
exports | 0x1ef6 - 0x2011 | 283 bytes | 14 count
elements | 0x2013 - 0x2025 | 18 bytes | 1 count
data count | 0x2027 - 0x2028 | 1 bytes | 1 count
code | 0x202c - 0x14154b | 1307935 bytes | 727 count
data | 0x14154f - 0x148ec9 | 31098 bytes | 2 count
custom "name" | 0x148ed2 - 0x1521a1 | 37583 bytes | 1 count
custom "producers" | 0x1521ae - 0x152256 | 168 bytes | 1 count
custom "target_features" | 0x152268 - 0x1522a6 | 62 bytes | 1 count
------ end module 0 -------------
module | 0x1522aa - 0x15ae70 | 35782 bytes | 1 count
------ start module 1 -------------
types | 0x1522b5 - 0x152338 | 131 bytes | 20 count
imports | 0x15233b - 0x152b4a | 2063 bytes | 41 count
functions | 0x152b4c - 0x152b92 | 70 bytes | 69 count
globals | 0x152b94 - 0x152ba4 | 16 bytes | 3 count
exports | 0x152ba7 - 0x152caa | 259 bytes | 17 count
code | 0x152cae - 0x15881a | 23404 bytes | 69 count
custom "name" | 0x158822 - 0x15ae21 | 9727 bytes | 1 count
custom "producers" | 0x15ae2d - 0x15ae70 | 67 bytes | 1 count
------ end module 1 -------------
module | 0x15ae73 - 0x15bb93 | 3360 bytes | 1 count
------ start module 2 -------------
types | 0x15ae7d - 0x15aeea | 109 bytes | 15 count
functions | 0x15aeec - 0x15af17 | 43 bytes | 42 count
tables | 0x15af19 - 0x15af1e | 5 bytes | 1 count
exports | 0x15af21 - 0x15aff5 | 212 bytes | 43 count
code | 0x15aff8 - 0x15b233 | 571 bytes | 42 count
custom "producers" | 0x15b23f - 0x15b263 | 36 bytes | 1 count
custom "name" | 0x15b26b - 0x15bb93 | 2344 bytes | 1 count
------ end module 2 -------------
module | 0x15bb96 - 0x15bd92 | 508 bytes | 1 count
------ start module 3 -------------
types | 0x15bba0 - 0x15bc0d | 109 bytes | 15 count
imports | 0x15bc10 - 0x15bd12 | 258 bytes | 43 count
elements | 0x15bd14 - 0x15bd44 | 48 bytes | 1 count
custom "producers" | 0x15bd50 - 0x15bd74 | 36 bytes | 1 count
custom "name" | 0x15bd7b - 0x15bd92 | 23 bytes | 1 count
------ end module 3 -------------
core instances | 0x15bd94 - 0x15bd98 | 4 bytes | 1 count
component alias | 0x15bd9a - 0x15bdac |
Ralph (Mar 11 2024 at 13:14):and llvm-dwarfdump sure doesn't know anything about components.
Ralph (Mar 11 2024 at 13:14):squillace@idiopath:~/work/hello-wasi-http-go$ dump -debug-line main.component.wasm
error: main.component.wasm: invalid version number: 65549
Ralph (Mar 11 2024 at 13:35):Scott Waye said:
I see, its compoents. I've not got there yet, but I will also be looking at it when I can get https://github.com/bytecodealliance/wasmtime/pull/8055 into a fit state.
This kind of seems like what https://github.com/WebAssembly/tool-conventions/issues/133 was for. Am I misunderstanding? Or, if I plow through the associated document, that seems more like keeping the debug sections but pointing to another location for the .dwo. How should I think about the two approaches for side-debugging symbols?
Last updated: Nov 22 2024 at 16:03 UTC