Some interesting findings about Chicago Build 73(f/g)

[roytam1]

While I'm trying to make my GreenPad text editor working on various windows builds, I found something interesting about Chicago Build 73:

The PE loader of Chicago Build 73(f/g) relies on "Section size in file" in ".bss" section.
normally modern linkers will put a 0 in this field, and normally PE loader will calculate the real size of this section in memory and give correct permissions for read and write.
but Chicago Build 73 doesn't do this but relies on "Section size in file" field instead, while modern linkers put a 0 to this field, Chicago Build 73 will simply not giving this section correct permissions.
As a result, program can't write to its global variables and causing a Page Fault in kernel32.dll every time it tries to access its global variables.

For a fix, I used PE Editor that comes from LordPE suite to edit .bss section header. I also used IDA to further verify the size of .bss section for me to fill the value (with rounding up to next 512 bytes)
and finally my binary is working in Chicago Build 73.

[LangsamSpieler]

Build 73 (Without Letter at end) is not a Real Build of Microsoft Chicago. See it on BetaWiki.
73f and 73g are the real ones.

[roytam1]

Build 73 (Without Letter at end) is not a Real Build of Microsoft Chicago. See it on BetaWiki.
73f and 73g are the real ones.

the behavior is common for 73f and 73g.
I can suppose it is common for all 73* builds.

[roytam1]

Just find that I don't really need to calculate `SizeOfRawData` by myself, I can just copy `IMAGE_OPTIONAL_HEADER.SizeOfUninitializedData` to it and done!
updated my little PE header utility for this: https://github.com/roytam1/rtoss/commit ... 88fe185f59

[DiskingRound]

Build 73 (Without Letter at end) is not a Real Build of Microsoft Chicago. See it on BetaWiki.
73f and 73g are the real ones.

73 must have existed. If 34 and 81 exist, there is definitely a 73, then recompiled into 73c, 73d, 73e, 73f, 73g, etc. to fix minor bugs.

[Battler]

Not necessarily, especially considering that the only difference between 73f and 73g is that the former is internal while the latter is public.

[roytam1]

Just find that I don't really need to calculate `SizeOfRawData` by myself, I can just copy `IMAGE_OPTIONAL_HEADER.SizeOfUninitializedData` to it and done!
updated my little PE header utility for this: https://github.com/roytam1/rtoss/commit ... 88fe185f59

and off-topic: and this reveals a bug in ReactOS' PE loader component. When `.bss` section's `SizeOfRawData` is not 0, program/library doesn't work correctly.

EDIT: it should be ntoskrnl/mm, created PR for it: https://github.com/reactos/reactos/pull/5002