Remove unnecessary computed goto

[pguyot]

The computed goto is a GCC extension that isn't implemented by clang when targeting wasm. In some cases the compiler failed to optimize it.

These changes are made under both the "Apache 2.0" and the "GNU Lesser General
Public License 2.1 or later" license terms (dual license).

SPDX-License-Identifier: Apache-2.0 OR LGPL-2.1-or-later

[bettio]

Change looks good, but CI fails with:

[100%] Built target test_suite_psa_crypto.pbkdf2
[100%] Built target test_suite_psa_crypto
In file included from /__w/AtomVM/AtomVM/src/libAtomVM/context.c:50:
/__w/AtomVM/AtomVM/src/libAtomVM/opcodesswitch.h:1875:14: error: WebAssembly hasn't implemented computed gotos
 1875 | HOT_FUNC int scheduler_entry_point(GlobalContext *glb)
      |              ^
/__w/AtomVM/AtomVM/src/libAtomVM/opcodesswitch.h:1875:14: error: WebAssembly hasn't implemented computed gotos
fatal error: error in backend: Cannot select: 0x5571b808bc40: i32 = BlockAddress<@scheduler_entry_point, %6247> 0
In function: scheduler_entry_point
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:

Are we missing anything?

[bettio]

It might be this one:

2849                     if (needs_to_wait) {
2850 #pragma GCC diagnostic push
2851 #pragma GCC diagnostic ignored "-Wpedantic"
2852                         ctx->restore_trap_handler = &&wait_timeout_trap_handler;
2853 #pragma GCC diagnostic pop
2854                         SCHEDULE_WAIT(mod, saved_pc);
2855                     } else {
2856                         JUMP_TO_ADDRESS(mod->labels[label]);
2857                     }
2858                 #endif

[pguyot]

I think recent version of emscripten no longer tolerates this pattern. This PR now completely gets rid of it.

[bettio]

Looks good to me, I'll merge it as soon as CI is green again.

By the way, I'm not understanding why this problem appeared only here and not on main branch.

[pguyot]

I think it's an emscripten image update thing and we didn't run CI on emscripten recently.

[bettio]

CI is failing since valgrind is detecting an uninitialised value.

[...]
add:
==16441== Conditional jump or move depends on uninitialised value(s)
==16441==    at 0x11D114: scheduler_entry_point (opcodesswitch.h:1925)
==16441==    by 0x11D015: context_execute_loop (opcodesswitch.h:1867)
==16441==    by 0x118AE4: test_atom (test.c:632)
==16441==    by 0x118D54: test_module_execution (test.c:682)
==16441==    by 0x118F12: test_modules_execution (test.c:738)
==16441==    by 0x1192E7: main (test.c:827)
==16441== 
add:OK
[...]
state_test2:
==16441== Thread 2:
==16441== Conditional jump or move depends on uninitialised value(s)
==16441==    at 0x11D114: scheduler_entry_point (opcodesswitch.h:1925)
==16441==    by 0x23C8BF: scheduler_thread_entry_point (smp.c:75)
==16441==    by 0x4A5B608: start_thread (pthread_create.c:477)
==16441==    by 0x4B95352: clone (clone.S:95)
==16441== 
state_test2:OK
[...]

I think the issue is waiting_with_timeout being not initialized.

[bettio]

So I agree with this change: there is no reason to use computed gotos. Still what was happening doesn't make sense:

CI in main is still successful with the same emsdk container that was failing on this branch:

main:

When it was failing here: