Re: Assignment to array/record constants.

[Gaius Mulley]

Guy <guyfieri@getbackinthe.kitchen> writes:

> Hello again Gaius, due to my lack of experience with mailing lists I'm afraid 
> I just recently realized that when attempting to reply to your response to 
> me, I accidentally replied directly to your personal email instead of the 
> mailing list. Due to my email address's domain name, I think that it probably 
> ended up in your junk/spam folder, my mistake.
>
> However since then I have installed Debian and was indeed able to compile the 
> project mostly without issue, so in hindsight it seems that the latest GCC 
> sources just aren't too well suited for macOS (though I'm sure there were 
> quite a few things I was missing on my end).
>
> So today, the issue/question I had is a bit different and instead related to 
> the actual compiler frontend's behavior. One of the first things I happened 
> to try after the successful compilation was some features that I made 
> frequent use of in other languages I have used, that being array/record 
> constants. I saw from a Modula-2 ISO reference that the language also 
> supported this via what it calls
> "constant constructors". Just to see how the compiler would respond, I tried 
> assigning a value to some test constants I made, but to my surprise it didn't 
> complain and actually replaced its value with the new one from the program 
> body.
>
> More specifically, assignments to the constants in the body of the program 
> via a constant constructor ("constant := TypeName{...}") replaced the current 
> value, but assignments to individual components of the array/record didn't 
> actually have an effect on the values. Running the test program in GDB. I 
> noticed the main procedure of executable programs seemed to be named
> _M2_<program_name>_init, and was wondering if all array/record constructors 
> (whether for a constant or not), were being moved into some sort of 
> pre-"main" area of code. When stepping through the program, the 
> new/additional assignments to the constant identifiers via constructors don't 
> appear in the places they do in the source.

Hello Guy,

a small followup to say that the bug has been fixed in the git repro
with an improved error location and message:

$ gm2 another.mod 
another.mod:8:17: error: In program module ‘another’: cannot assign a constant 
designator expression with an expression ‘2.0’
    8 |    Fractions[2] := 2.0
      |    ~~~~~~~~~~~~~^~~~~~
$ gm2 constants.mod 
constants.mod:34:14: error: In program module ‘Constants_Example’: cannot 
assign a constant designator ‘Test_Point’ with an expression constructor
   34 |   Test_Point := TPoint{100.0, 80.0};
      |   ~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~
constants.mod:62:14: error: cannot assign a constant designator ‘Test_Color’ 
with an expression constructor
   62 |   Test_Color := TColor{0, 0, 0};
      |   ~~~~~~~~~~~^~~~~~~~~~~~~~~~~~
constants.mod:64:25: error: cannot assign a constant designator expression with 
an expression ‘54’
   64 |   Test_Color[Color_Red] := 54;
      |   ~~~~~~~~~~~~~~~~~~~~~~^~~~~
constants.mod:65:27: error: cannot assign a constant designator expression with 
an expression ‘97’
   65 |   Test_Color[Color_Green] := 97;
      |   ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~
constants.mod:66:26: error: cannot assign a constant designator expression with 
an expression ‘182’
   66 |   Test_Color[Color_Blue] := 182;
      |   ~~~~~~~~~~~~~~~~~~~~~~~^~~~~~


I see that a similar test with constant record also exhibits a failure -
so will fix this using a similar route (designator record vs designator
array)

regards,
Gaius