1. Signed Hex Values
- Posted by petelomax Sep 22, 2013
- 1700 views
A proposed change, in documentation form.
Signed Hex Values
By default hexadecimal values are unsigned. Use "#-" to define signed hexadecimal values, eg:
#FFFFFFFF -- +4294967295 #-FFFFFFFF -- -1 #80070057 -- +2147942487 #+80070057 -- +2147942487 #-80070057 -- -2147024809 -- (the desired value for E_INVALIDARG) -#80070057 -- -2147942487
Note that "#-" is a compiler directive, rather than a unary minus as just seen in that last example. You can also code "#+" for alignmnent purposes, which behaves identically to "#". Some more examples:
#00000001 -- +1 #+00000001 -- also +1 #-00000001 -- also +1 (!! make sure you understand why this is plus 1 !!) -#00000001 -- -1 #-FF -- -1 #-00FF -- +255 (make sure you properly sign-extend short forms when editing them) if offset>=#-80 and offset<=#7F then use_short_form() end if if offset>=-128 and offset<=127 then use_short_form() end if -- (same as above)
The #- directive triggers a compilation error if the following hexadecimal constant is not 2, 4, or 8 characters, or 16 in 64-bit code.
Comments welcome.
Pete
2. Re: Signed Hex Values
- Posted by DerekParnell (admin) Sep 23, 2013
- 1596 views
A proposed change ...
Signed Hex Values
Why is this a good idea? What's the advantage, etc ...
If adopted, following the same logic, we'd also have to allow the following combinations ...
+#-FFFFFFFF -- -1 -#-FFFFFFFF -- +4294967295 +#+FFFFFFFF -- +4294967295 -#+FFFFFFFF -- -1
The #- directive triggers a compilation error if the following hexadecimal constant is not 2, 4, or 8 characters, or 16 in 64-bit code.
Why have this restriction?
And what about binary literals and hex values inside strings?
3. Re: Signed Hex Values
- Posted by petelomax Sep 23, 2013
- 1550 views
A proposed change ...
Signed Hex Values
Why is this a good idea? What's the advantage, etc ...
Things like E_INVALIDARG (and many many friends). I get a negative HRESULT, I know E_INVALIDARG is 0x80070057, but erm, -2147024809? What else might you suggest?
If adopted, following the same logic, we'd also have to allow the following combinations ...
+#-FFFFFFFF -- -1 -#-FFFFFFFF -- +4294967295 +#+FFFFFFFF -- +4294967295 -#+FFFFFFFF -- -1
Yes, I had spotted that potential for code obfuscation
The correct values would be:
+#-FFFFFFFF -- -1 -#-FFFFFFFF -- +1 +#+FFFFFFFF -- +4294967295 -#+FFFFFFFF -- -4294967295
which just goes to show it would be quite good for that. I don't really see what the problem with any of that would be though.
The #- directive triggers a compilation error if the following hexadecimal constant is not 2, 4, or 8 characters, or 16 in 64-bit code.
Why have this restriction?
I did't think it would be sensible for the compiler to effectively decide on a quite arbitrary basis that say bit 23 was the sign bit?
And what about binary literals and hex values inside strings?
I'm not seeing the same kind of comparison requirement there, so no.
4. Re: Signed Hex Values
- Posted by DerekParnell (admin) Sep 23, 2013
- 1618 views
It turns out that I misread your initial examples.
I'm now totally confused about what you are trying to suggest. Your examples seem counter-intuitive and would need a lot of explaining to make sure people understood its intent. And that is an alarm bell for me. In fact, I don't understand some of them at all. Sorry.
Can you explain what is the problem that this idea is attempting to solve?
5. Re: Signed Hex Values
- Posted by petelomax Sep 23, 2013
- 1567 views
Ah, nevermind. I was doing
push eax fild dword[esp] add esp,4then checking for error using hResult<0 but not having any of the right values to compare against. If instead I
push dword 0 push eax fild qword[esp] add esp,8and check for error using and_bits(hResult,#80000000) then (eg) #800704C7 is fine.
Pete
6. Re: Signed Hex Values
- Posted by mattlewis (admin) Sep 24, 2013
- 1513 views
...checking for error using hResult<0 but not having any of the right values to compare against. If instead I... and check for error using and_bits(hResult,#80000000) then (eg) #800704C7 is fine.
Yes, that's how you're supposed to check for errors. It's what I do in EuCOM:
--/topic Errors --/func is_error( atom hresult ) --/ret Boolean: 1 if error, 0 if not error --Typically, an hresult (the standard return value from a COM function call) will --be 0 if no error, and non-zero if an error occurs. However, there are some values that --are non-zero that are 'qualified' successes. Critical errors, as they are called, can --be detected by using a mask of #80000000, which is what this funcion does. public function is_error( atom hresult ) return and_bits( hresult, #80000000 ) end function
Matt
