Strange memory access

[rfalke]

Subject: https://github.com/rfalke/decompiler-subjects/tree/master/from_holdec/dmi/value_prop/ia32_elf

Output:

Eq_369 advanced_2_mem_values_can_be_propagated_for_non_write_calls(char * dwArg04)
{
	g_t804A02C = 0x01;
	return strlen(dwArg04) + Mem16[0x0804A02C<p32>:word32];
}

Expected:

Eq_369 advanced_2_mem_values_can_be_propagated_for_non_write_calls(char * dwArg04)
{
	g_t804A02C = 0x01;
	return strlen(dwArg04) + g_t804A02C;
}

I'm also not sure what Mem16 should be. The read access is 32bit.

[uxmal]

What you are seeing is a crash in the type inference part of Reko. Trying to recover, it emits in desperation the low-level RTL representation of a memory access. Memory accesses in Reko are expressed as a tuple (MemorySpace, effective address) -- ignoring the x86-specific SegmentedMemoryAccesses). Memory spaces are SSA numbered, just like identifiers are. A future version will be able to take advantage of SSA-numbered memory spaces to make decisions about memory accesses aliasing each other.

The crash should ofc never happen. I'm assigning this to myself for investigation.

[rfalke]

Output of version 0.11.4.0-931ca7d:

Eq_360 advanced_2_mem_values_can_be_propagated_for_non_write_calls(char * dwArg04)
{
	global_var = 0x01;
	return (word32) strlen(dwArg04) + global_var;
}

Fixed -> close issue.