Resumption of Work on M2C

[trijezdci]

Work on M2C Resumed

The M2C project was on hold but I have now resumed working on it.

The Backstory

I had the front end working a few years back and it was quite heavily tested by many folks on the Usenet m2 group. The compiler was supposed to be used as a bootstrap and porting tool for our revised Modula-2 language project (M2R10), hence the implementation in C, translation to C and compilation via C. The aim was always to generate C source code that is authentic and human readable, that is to say, it should look mostly as if it had been written by a human programmer.

However, back then many people wanted unrestricted use of lowlines _ in Modula-2 identifiers and the generated C code was to be in MACRO_CASE for constants and snake_case for everything else. These two aims made collision free identifier translation from Modula-2 to C far too complex and the implementation unwieldy. This ultimately put the entire project on indefinite hold.

Instead, I manually transcribed the code to Modula-2, using only a subset available in both PIM and ISO Modula-2 so that the code wouldn't depend on any particular Modula-2 compiler. That project is also on Github:

https://github.com/m2sf/m2bsk

Furthermore, @viegasfh is working on the .NET hosted C# version which will be generating code for .NET.

https://github.com/m2sf/m2sharp

Unfortunately, the two Modula-2 compilers we were using for the Modula-2 hosted version are stuck with 32-bit. In the case of MOCKA it would require very serious work to update it to 64-bit because the backend was generated by a closed source tool, thus an entire new backend would have to be developed from scratch. In the case of p1, the compiler isn't open source and only the older 32-bit version is free of charge. And with GM2, there are always issues getting it to work on the various platforms we want to target.

Before this background, I have taken another look at the identifier conversion issue that put M2C on hold. I have now changed the lexer so that lowline use in Modula-2 identifiers is restricted to non-leading, non-trailing and non-consecutive occurrences. Then I designed a much simpler name translation scheme that assumes that people will by and large stick to the Modula-2 naming convention where module, type and procedure identifiers are in TitleCase, variable and function identifiers in camelCase and constant identifiers consistently in one of those two.

https://github.com/trijezdci/m2c/wiki/Identifier-Mapping

I still need to implement this and completely rewrite m2-identconv.c.

After implementing the name conversion, there is still some other work to be done, but the compiler is not all that far from completion. I will first focus on PIM3 and PIM4 and may initially omit some semantic checks, in order to get a usable translator from PIM3/4 to C because that could then be used to continue work on the Modula-2 hosted M2BSK compiler.

benjamin