[Coco] CC issue

[Walter Zambotti]

Jeff

I'm using the c.comp source from the MC cross compiler which contains a 
lot of C code.  I can only assume the OS9 c.pass1 and c.pass2 are 
similar because they exhibit the same issue.

Anyway debugging through c.comp it appears that preprocessor statements 
are handled separately from C statements.  Moving the logic around 
appears complicated due to internal structures that are not well documented.

Even if I managed to find a simple fix for c.comp trying to find the 
respective fix in the c.pass1/2 is a different story as I have no source 
(unless there is source).

Walter

On 11/10/19 4:35 AM, Jeff Teunissen wrote:
> Of course, I am speaking as someone looking at source, not trying to
> grovel through a binary. build.c (comp_03.c in the decompiled source)
> is PROBABLY where you should look (it is responsible for building the
> compiler's expression trees), but possibly tranexp.c (AKA comp_13.c),
> which does expression translation.
>
> On Sat, Nov 9, 2019 at 2:05 AM Jeff Teunissen <deek at d2dc.net> wrote:
>> Sounds like it's actually two bugs, and both should be fairly simple.
>>
>> Firstly, preprocessor statements that actually make it to the compiler
>> need to be considered new statements.
>> Secondly, the end of a block should complete a statement.
>>
>> On Fri, Nov 8, 2019 at 12:52 AM Walter Zambotti <zambotti at iinet.net.au> wrote:
>>> I found a small issue in both c.pass1/2 and o c_comp.
>>>
>>>
>>>
>>> In the following demonstrative code (which is not meant to be logical):
>>>
>>>
>>>
>>> main()
>>>
>>> {
>>>
>>>    int I;
>>>
>>>    while(i++) { i--;}
>>>
>>> #asm
>>>
>>>    clra
>>>
>>>    clrb
>>>
>>> #endasm
>>>
>>>    i = i + 1;
>>>
>>> }
>>>
>>>
>>>
>>> The asm code is incorrectly placed inside the while loop.
>>>
>>>
>>>
>>> This occurs because the compiler is looking for a semicolon after the loop
>>> block.  Which of course is not required and is tolerated.
>>>
>>> However while it is looking for the semicolon if it comes across any
>>> comments it outputs them immediately doing so before it has
>>> finalised the output for the loop.
>>>
>>>
>>>
>>> This issue does effect program execution.
>>>
>>>
>>>
>>> The easy work around is to place an extraneous semicolon after the loop
>>> block like so:
>>>
>>>
>>>
>>>    while(i++) { i--;};
>>>
>>> or
>>>
>>> while(i++) i++;;
>>>
>>>
>>>
>>> This forces the compiler to place the asm code after the loop as expected.
>>>
>>>
>>>
>>> This occurs for any statement block that contains exit logic such as:
>>>
>>>
>>>
>>> While, for, switch
>>>
>>>
>>>
>>> This also affects C lines added as comments to the assembly.  The C comment
>>> lines appear out of context to their respective asm.
>>>
>>>
>>>
>>> This last issue does not effect program execution but makes it difficult to
>>> use the assembly listing to debug your code.
>>>
>>>
>>>
>>> I have traced through c.comp extensively and the fix would not be easy.
>>>
>>>
>>>
>>> I did however manage to add a fix to c.prep so that it outputs a semicolon
>>> before any asm code.  This seems to fix the
>>> issue but would need extensive testing.
>>>
>>>
>>>
>>> Walter
>>>
>>>
>>> --
>>> Coco mailing list
>>> Coco at maltedmedia.com
>>> https://pairlist5.pair.net/mailman/listinfo/coco