modaccent long type conversation errors

[HexDecimal]

I've noticed a ton of conversion errors in the modaccent sources. These sources seem to insist on casting long types to long* pointers and back again. On most x64 systems the pointer will be larger than the integer.

This can generally be fixed by replacing long with intptr_t.

[rmtew]

Currently playing with this.

[rmtew]

This however, is fixing the output of generated code.. better would be to regenerate them with better tools, but whether we had the source files I do not know.

[HexDecimal]

Generated code explains is why it looks so terrible. I'm not sure where it's generated from.

[rmtew]

Possibly it came wholesale as part of the ACCENT distribution and was used with a bit of wrapping by Julian. Possibly he lexed and bisoned it together ACCENT. Either way it is what it is.

[rmtew]

I think we have our answer :-)

https://github.com/viegasfh/accent/tree/master/accent

[rmtew]

I think the right path forward is working out how to build a modern accent. I've fixed the intptr_t occurences and it goes from crashing to existing with some maxint looking exit code.

[rmtew]

1. Get some linux adjacent environment with modern flex and bison.
2. Compile Gentle-97.
3. Edit and run accent build script.

[rmtew]

Rerunning accent or gentle fixes none of the core problems here.

• Gentle and accent have the long * baked in.
• Functions are called without definitions, both in the same file and externally. Arguments default to int and get clipped if actually 64 bit.

This is the same whether using the new cleaner accent output or the existing modaccent code. At some point only compiling it as 32-bit and running that, ignoring all the warnings and so on seems like the reasonable outcome.

[HexDecimal]

It's difficult to find information on this. Which parts of the project rely on this compiler? Is this for the files in lib/? Or for the ones in lang/?

[HexDecimal]

Grammar.acc, the .acc is for ACCENT. So now I need to figure out what Tokens.lex is. I somewhat understand, but it's very convoluted.

[rmtew]

Here's me trying to write it as simply as possible. As it is convoluted this might be wrong.

1. We run flex on Tokens.lex to get the IncursionScript lexer code called by the accent parser/compiler.
2. We run modaccent on Grammar.acc to get the IncursionScript parser/compiler.
3. We build this into Incursion.exe debug builds.
4. We execute a debug Incursion.exe with -compile to create the module.

I've never bothered looking at this whole side of things outside of bug fixing in the grammar because IMO it is a time sink. What I would personally do is refactor the module compiling out into a separate executable and remove it from Incursion.exe. And continue to ignore this whole part of it and focus time on user accessible things.

[HexDecimal]

It does look like messing with this could become a time sink. I should just be happy that the automated builds actually work right now.