cc65/Contributing.md

Contributing to cc65
====================

This document contains all kinds of information that you
should know if you want to contribute to the cc65 project.
Before you start, please read all of it.  If something is not
clear to you, please ask - this document is an ongoing effort
and may well be incomplete. 

Also, before you put a lot of work into implementing
something you want to contribute, please get in touch with
one of the developers and ask if what you are going to do is
actually wanted and has a chance of being merged.  Perhaps
someone else is already working on it, or perhaps what you
have in mind is not how we'd expect it to be - talking to us
before you start might save you a lot of work in those cases.

(''Note:'' The word "must" indicates a requirement.  The word
 "should" indicates a recomendation.)

*this is work in progress and is constantly updated - if in
doubt, please ask*

# Generally

* You must obey these rules when contributing new code or
  documentation to cc65. We are well aware that not all
  existing code may respect all rules outlined here - but this
  is no reason for you not to respect them.
* One commit/patch/PR per issue. Do not mix several things
  unless they are very closely related.
* Sometimes when you make a PR, it may break completely
  unrelated tests. However, any PR is expected to merge
  cleanly with no failures.  That means in practise that you
  are expected to fix/update the failing tests if required -
  for example this might be needed if you make changes to the
  compiler that changes the format of error- or warning
  messages.  In that case you might have to update some
  reference files in the testbench.  Obviously still check if
  that is actually the right thing to do. ;)

# Codestyle rules

## All sources

### Line endings

All files must only contain Unix style 'LF' line endings.
Please configure your editors accordingly.

### TABs and spaces

This is an ongoing controversial topic - everyone knows
that.  However, the following is how we do it :)

* TAB characters must be expanded to spaces.
* 4 spaces per indention level (rather than 8) are
  preferred, especially if there are many different levels.
* No extra spaces at the end of lines.
* All text files must end with new-line characters.  Don't
  leave the last line "dangling".

The (bash) scripts used to check the above rules can be
found in ```.github/check```.  You can also run all checks
using ```make check```.

### Identifiers and symbol names

The C Standard defines certain identifiers and symbol names,
which we can not use in our code.  Since it is not always
obvious which parts of the library code will actually end up
in a linked program, the following applies to ALL of the
library.

Any non standard identifier/symbol/function that is exported
from source files, or appears in header files:

* must not be in the "_symbol" form in C, or "__symbol" form
  in assembly,
* must start with (at least) two (C Code) or three (assembly
  code) underscores, unless the symbol appears in a non
  standard header file.

This is likely more than the standard dictates us to do -
but it is certainly standard compliant - and easy to
remember.

Also see the discussion in
https://github.com/cc65/cc65/issues/1796

### Miscellaneous

* 80 characters is the desired maximum width of files.  But,
  it isn't a "strong" rule; sometimes, you will want to type
  longer lines, in order to keep the parts of expressions or
  comments together on the same line.
* You should avoid typing non-ASCII characters.
* If you change "normal" source code into comments, then you
  must add a comment about why that code is a comment.
* When you want to create a comment from several lines of
  code, you should use preprocessor lines, instead of ```/*
  */``` or "```;```".  Example:

~~~C
#if 0
  one (); two ();
  three = two () + one ();
#endif
~~~

* You should type upper case characters for hex values.
* When you type zero-page addresses in hexadecimal, you
  should type two hex characters (after the hex prefix).
  When you type non-zero-page addresses in hex, you should
  type four hex characters.
* When you type lists of addresses, it is a good idea to
  sort them in ascending numerical order.  That makes it
  easier for readers to build mental pictures of where things
  are in an address space.  And, it is easier to see how big
  the variables and buffers are.  Example:

~~~asm
xCoord := $0703
yCoord := $0705 ; (this address implies that xCoord is 16 bits)
cmdbuf := $0706 ; (this address implies that yCoord is 8 bits)
cmdlen := $0786 ; (this address implies that cmdbuf is 128 bytes)
color  := $0787
~~~

## C sources

The following is still very incomplete - if in doubt please
look at existing sourcefiles and adapt to the existing style.

Your files should generally obey the C89 standard, with a
few C99 things (this is a bit similar to what cc65 itself
supports). The exceptions are:

* Use stdint.h for variables that require a certain bit size
* In printf-style functions use the PRIX64 (and similar)
  macros to deal with 64bit values (from inttypes.h) This
  list is not necessarily complete - if in doubt, please ask.
* We generally have a "no warnings" policy
* Warnings must not be hidden by using typecasts - fix the
  code instead
* The normal indentation width should be four spaces.
* You must use ANSI C comments (```/* */```); you must not
  use C++ comments (```//```).
* When you add functions to an existing file, you should
  separate them by the same number of blank lines that
  separate the functions that already are in that file.
* All function declarations must be followed by a comment
  block that tells at least briefly what the function does,
  what the parameters are, and what is returned. This comment
  must sit between the declaration and the function body, like
  this:

~~~C
int foo(int bar)
/* Add 1 to bar, takes bar and returns the result */
{
    return bar + 1;
}
~~~

* When a function's argument list wraps around to a next
  line, you should indent that next line by either the
  normal width or enough spaces to align it with the arguments
  on the previous line.
* All declarations in a block must be at the beginning of
  that block.
* You should put a blank line between a list of local
  variable declarations and the first line of code.
* Always use curly braces even for single statements after
  ```if```, and the single statement should go into a new
  line.
* Use "cuddling" braces, ie the opening brace goes in the
  same line as the ```if```:

~~~C
if (foo > 42) {
    bar = 23;
}
~~~

* Should the ```if``` statement be followed by an empty
  conditional block, there should be a comment telling why
  this is the case:

~~~C
if (check()) { /* nothing happened, do nothing */ }
~~~

* You must separate function names and parameter/argument
  lists by one space.
* When declaring/defining pointers, you must put the
  asterisk (```*```) next to the data type, with a space
  between it and the variable's name.  Examples:

~~~C
int* namedPtr[5];
char* nextLine (FILE* f);
~~~

### Header files

Headers that belong to the standard library (libc) must
conform with the C standard. That means:

* All non standard functions, or functions that only exist
  in a certain standard, should be in #ifdefs
* The same is true for macros or typedefs.
  You can refer to Annex B of the ISO C99 standard
  ([here](https://www.open-std.org/jtc1/sc22/wg14/www/docs/n1256.pdf)
  is the draft).  Example:

~~~C
#if __CC65_STD__ == __CC65_STD_C99__ /* stuff that only exists in C99 here */
#endif
#if __CC65_STD__ == __CC65_STD_CC65__ /* non standard stuff here */
#endif </pre>
~~~

## Assembly sources

* Opcode mnemonics must have lower-case letters.  The names
  of instruction macroes may have upper-case letters.
* Opcodes must use their official and commonly used
  mnemonics, ie 'bcc' and 'bcs' and not 'bgt' and 'blt'.
* Hexadecimal number constants should be used except where
  decimal or binary numbers make much more sense in that
  constant's context.
* Hexadecimal letters should be upper-case.
* When you set two registers or two memory locations to an
  immediate 16-bit zero, you should use the expressions
  ```#<$0000``` and ```#>$0000``` (they make it obvious where
  you are putting the lower and upper bytes).
* If a function is declared to return a char-sized value, it
  actually must return an integer-sized value.  (When cc65
  promotes a returned value, it sometimes assumes that the value
  already is an integer.)
  This must be done in one of the following ways:

~~~asm
    lda #RETURN_VALUE
    ldx #0 ; Promote char return value

    ; If the value is 0, you can use:
    lda #RETURN_VALUE
    .assert RETURN_VALUE = 0
    tax

    ; Sometimes jumping to 'return 0' could save a byte:
    .assert RETURN_VALUE = 0
    jmp return 0
~~~

* Functions, that are intended for a platform's system
  library, should be optimized as much as possible.
* Sometimes, there must be a trade-off between size and
  speed.  If you think that a library function won't be used
  often, then you should make it small.  Otherwise, you should
  make it fast.
* Comments that are put on the right side of instructions
  must be aligned (start in the same character columns).
* Assembly source fields (label, operation, operand,
  comment) should start ''after'' character columns that are
  multiples of eight (such as 1, 9, 17, 33, and 41).
  
## LinuxDoc sources

* TAB characters must be expanded to spaces.
* All text files must end with new-line characters.  Don't
  leave the last line "dangling".
* 80 characters is the desired maximum width of files.
* You should avoid typing non-ASCII characters.

* You should put blank lines between LinuxDoc sections:
* Three blank lines between ```<sect>``` sections.
* Two blank lines between ```<sect1>``` sections.
* One blank line between other sections.

# Library implementation rules

* By default the toolchain must output a "standard" binary
  for the platform, no emulator formats, no extra headers
  used by tools.  If the resulting binaries can not be run as
  is on emulators or eg flash cartridges, the process of
  converting them to something that can be used with these
  should be documented in the user manual.
* Generally every function should live in a seperate source
  file - unless the functions are so closely related that
  splitting makes no sense.
* Source files should not contain commented out code - if
  they do, there should be a comment that explains why that
  commented out code exists.

# Makefile rules

* Makefiles must generally work on both *nix (ba)sh and
  windows cmd.exe.
* Makefiles must not use external tools that are not
  provided by the cc65 toolchain itself.

The only exception to the above are actions that are exclusive
to the github actions - those may rely on bash and/or linux tools.

# Documentation rules

## User manual (LinuxDoc)

* This is the primary documentation.

## Wiki

* The Wiki is strictly for additional information that does
  not fit into the regular user manual (LinuxDoc).  The wiki
  must not duplicate any information that is present in the
  user manual.

# Roadmap / TODOs / open ends

## Documentation

* The printf() family of functions does not completely
  implement all printf() modifiers and does not behave as
  expected in some cases - all this should be documented in
  detail.

## Compiler

* We need a way that makes it possible to feed arbitrary
  assembler code into the optimzer, so we can have proper
  tests for it.

### Floating point support

The first step is implementing the datatype "float" as IEEE
754 floats.  Help welcomed!

* WIP compiler/library changes are here:
  https://github.com/cc65/cc65/pull/1777

## Library

Some name clashes need to be resolved.  Please see the
[detailed list of name clashes](libsrc/NameClashes.md).

## Test suite

* Specific tests to check the optimizer (rather than the code
  generator) are needed.
* We need more specific tests to check standard conformance
  of the library headers.