Fix auto reload scripts on external change

[cptchuckles]

fixes #49298

[KoBeWi]

So I tested this and it doesn't seem to work? I tested with a simple script

extends Node
@export var test = 4

When I change value of test, it's supposed to be updated in the inspector, no? It doesn't happen.

[cptchuckles]

@KoBeWi Without this patch, the editor actually doesn't reload the script at all and the major problem is when you add or remove export variables; you would have to close and reopen the entire scene in order to see the most current exports. This does get fixed. As I discovered and wrote in the related issue #49298, the problem is that GDScriptCache::get_full_script() doesn't check file modified time before returning what's in the cache when the editor regains focus, so it would never know if the script changed since it was cached.

However, like you I noticed recently that half of the time, the editor appears to first save the old value in the .tscn as a property of the node (and the reset button appears next to the value in the inspector), and other times the editor wouldn't save the old value but instead populate it with the new value from the script as expected. I don't know why that happens now, though, nor especially why it's intermittent. Months ago, when I first opened this PR, it actually did update values and would never save old values in the scene file, but now it does, so I'll probably have to track that down too.

[cptchuckles]

I tracked the latest bug related to this (new defaults getting overridden by old defaults) down to 4e6efd1, which introduced C++ iterators. There were several places where E->get() was changed to just E where it shouldn't have been, one such occurrence being now fixed by #55350 which solves it. Too bad I didn't check the latest build before going on a bisect... lol

This patch is now once again working as expected on the latest revision. I still need a reviewer to tell me whether my solution is stupid or not though

[akien-mga]

Thanks!

[nathanfranke]

I get a SIGSEGV in my main project after this commit.

#0  0x000055555825084e in Map<StringName, GDScriptCodeGenerator::Address, Comparator<StringName>, DefaultAllocator>::_find (
    this=<error reading variable: Cannot access memory at address 0x7fffff7fefc8>, 
    p_key=<error reading variable: Cannot access memory at address 0x7fffff7fefc0>) at ./core/templates/map.h:299
#1  0x000055555824dc92 in Map<StringName, GDScriptCodeGenerator::Address, Comparator<StringName>, DefaultAllocator>::find (this=0x7fffff800748, 
    p_key=...) at ./core/templates/map.h:598
#2  0x000055555824c3fb in Map<StringName, GDScriptCodeGenerator::Address, Comparator<StringName>, DefaultAllocator>::operator[] (this=0x7fffff800748, 
    p_key=...) at ./core/templates/map.h:671
#3  0x0000555558239312 in GDScriptCompiler::_parse_expression (this=0x7fffff801b80, codegen=..., r_error=@0x7fffff80027c: OK, 
    p_expression=0x5555b4816b40, p_root=false, p_initializer=false, p_index_addr=...) at modules/gdscript/gdscript_compiler.cpp:226
#4  0x000055555823c49a in GDScriptCompiler::_parse_expression (this=0x7fffff801b80, codegen=..., r_error=@0x7fffff80027c: OK, 
    p_expression=0x5555b4816ca0, p_root=false, p_initializer=false, p_index_addr=...) at modules/gdscript/gdscript_compiler.cpp:670

[fire]

Can report this breaks our project. I recommend to revert this change. We have a large project and it broke.

Teammate.
"Generally using timestamps for change checking is problematic."