For debugging purposes, I have built a mixin that will, when declared inside a function, output code to the console that will reproduce the exact function call.

So, as an example, for the following function

int reproducible_function(int a, int b){
  mixin(OUTPUT_REPRO_CASE!reproducible_function);
  return a + b;
}

called in the following way:

reproducible_function(5, 9);

the mixin OUTPUT_REPRO_CASE will output the exact same call. I can then take that code, paste it into the tests and tadaaaa - I have a repro case.

The only problem I have with this is that you still need to give the function as a template argument to OUTPUT_REPRO_CASE. In other cases, I have gotten around this by giving __FUNCTION__ as the default argument to that template parameter, but that doesn't work here because the mixin expands to something that calls ParameterIdentifierTuple. ParameterIdentifierTuple does not take a string as an argument, but needs the actual symbol (or an alias) to be passed.

So what I am looking for then is the equivalent to __FUNCTION__ that evaluates to the actual symbol of the function instead of its name, so it can be used as a parameter to ParameterIdentifierTuple. With this, I could then change the call to just be:

int reproducible_function(int a, int b){
   mixin(OUTPUT_REPRO_CASE!());
   return a + b;
}

On Saturday, 23 October 2021 at 18:23:47 UTC, Simon wrote:
> So what I am looking for then is the equivalent to __FUNCTION__ that evaluates to the actual symbol of the function instead of its name, so it can be used as a parameter to ParameterIdentifierTuple.

You could use the following:

alias F = __traits(parent, {});

The lambda {} is a symbol inside the function. F will be its parent, which is the function itself.

On Saturday, 23 October 2021 at 18:36:27 UTC, Tim wrote:
> On Saturday, 23 October 2021 at 18:23:47 UTC, Simon wrote:
>> So what I am looking for then is the equivalent to __FUNCTION__ that evaluates to the actual symbol of the function instead of its name, so it can be used as a parameter to ParameterIdentifierTuple.
>
> You could use the following:
>
> alias F = __traits(parent, {});
>
> The lambda {} is a symbol inside the function. F will be its parent, which is the function itself.

So before I had OUTPUT_REPRO_CASE defined like this:

enum OUTPUT_REPRO_CASE(alias func) = "build the actual code stuff with"~fullyQualifiedName!func~" and so on";

And I tried to use your suggestion like this:

enum OUTPUT_REPRO_CASE(alias func = __traits(parent, {})) = "build the actual code stuff with"~fullyQualifiedName!func~" and so on";

Which doesn't work. In that case func seems to become the parent namespace. If I substitute fullyQualifiedName!func with fullyQualifiedName!(__traits(parent, {}))) I get a circular dependecy error.

How do I have to use this exactly?

On Saturday, 23 October 2021 at 18:56:48 UTC, Simon wrote:
> And I tried to use your suggestion like this:
>
> enum OUTPUT_REPRO_CASE(alias func = __traits(parent, {})) = "build the actual code stuff with"~fullyQualifiedName!func~" and so on";
>
> Which doesn't work. In that case func seems to become the parent namespace. If I substitute fullyQualifiedName!func with fullyQualifiedName!(__traits(parent, {}))) I get a circular dependecy error.
>
> How do I have to use this exactly?

This seems to work:

import std.traits, std.stdio;
enum OUTPUT_REPRO_CASE = q{
    string name = fullyQualifiedName!(__traits(parent, {}));
    writeln(name);
};
void test()
{
    mixin(OUTPUT_REPRO_CASE);
}
void main()
{
    test();
}

On Saturday, 23 October 2021 at 19:03:41 UTC, Tim wrote:
> On Saturday, 23 October 2021 at 18:56:48 UTC, Simon wrote:
>> And I tried to use your suggestion like this:
>>
>> enum OUTPUT_REPRO_CASE(alias func = __traits(parent, {})) = "build the actual code stuff with"~fullyQualifiedName!func~" and so on";
>>
>> Which doesn't work. In that case func seems to become the parent namespace. If I substitute fullyQualifiedName!func with fullyQualifiedName!(__traits(parent, {}))) I get a circular dependecy error.
>>
>> How do I have to use this exactly?
>
> This seems to work:
>
> import std.traits, std.stdio;
> enum OUTPUT_REPRO_CASE = q{
>     string name = fullyQualifiedName!(__traits(parent, {}));
>     writeln(name);
> };
> void test()
> {
>     mixin(OUTPUT_REPRO_CASE);
> }
> void main()
> {
>     test();
> }

Thanks for putting up with me! I tried a bunch and it seems like I left out too much code, because I can't get it working 100%. I didn't even know about the q{} syntax, so I didn't think the stuff I left out would matter, but when using this approach it apparently does. I am going to show the full code here, just so that there are no more suprises.

Sorry for this becoming a "please do my homework" kind of question. I tried for half an hour with no idea how to fix the errors I'm seeing. Something always fails.

The full implementation of OUTPUT_REPRO_CASE looked like this:


enum MAKE_FUNCTION_REPRODUCIBLE(alias func, s64 execution_count_to_log_reproduction_on = -1) =
  "static s64 execution_count = 0;
  if(context.settings.debug_.output_call_counts_for_reproducible_functions)
    dlog(\""~fullyQualifiedName!func~" call count: %lld\", execution_count);

  if(execution_count == "~execution_count_to_log_reproduction_on.stringof~"){
    dlog(generate_reproduction_for_current_function_call!(
      "~fullyQualifiedName!func~",
      ["~get_string_argument_list_of_function!([ParameterIdentifierTuple!func])~"])(
      context.temp_allocator,
      "~get_argument_list_of_function!([ParameterIdentifierTuple!func])~"));

      trigger_breakpoint();
}";

It calls the function

String generate_reproduction_for_current_function_call(alias func, string[] parameter_names, Parameter_Types...)(Allocator* allocator, Parameter_Types parameters);

that actually creates the code-string. I did a lot of trial and error getting the trait-calls to expand at the right time (become part of the mixin string or outside of it) and falling on my nose about when something is a type vs a string and so on, so this became really ugly. If you have a better suggestion I'm open to it.

I was testing this inside the function

Dynamic_Array!(Polygon) clip_polygons(Slice!Vector2 p0, Slice!Vector2 p1, bool return_difference_instead_of_union = true, Slice!(Slice!Vector2) p0_holes = Slice!(Slice!Vector2)(), bool debug_print = false);

in which the OUTPUT_REPRO_CASE expands to:

static s64 execution_count = 0;
  if(context.settings.debug_.output_call_counts_for_reproducible_functions)
    dlog("vbl.clip_polygons call count: %lld", execution_count);

  if(execution_count == -1L){
    dlog(generate_reproduction_for_current_function_call!(
      vbl.clip_polygons,
      ["p0", "p1", "return_difference_instead_of_union", "p0_holes", "debug_print"])(
      context.temp_allocator,
      p0, p1, return_difference_instead_of_union, p0_holes, debug_print));

      trigger_breakpoint();
}

When I try to change the whole thing to be a token string, execution_count_to_log_reproduction_on was suddenly undefined. I changed this to be an abominational mix of token string and string literal and then it sort of worked, except then the whole business of making the parameter names expand as actual tokens and not strings failed for some reason.

I am super confused (this is a reccuring theme with D compile time stuff on my part :/). Can you show me how this can work with your token string solution?

On Saturday, 23 October 2021 at 18:23:47 UTC, Simon wrote:
> For debugging purposes, I have built a mixin that will, when declared inside a function, output code to the console that will reproduce the exact function call.

Sounds like what you really want is

https://github.com/dlang/dmd/pull/13071

but the PR there is not merged :(

the other way though is the string mixin with the parent {} thing it is pretty ugly.

On Saturday, 23 October 2021 at 19:52:19 UTC, Simon wrote:
>
> Thanks for putting up with me! I tried a bunch and it seems like I left out too much code, because I can't get it working 100%. I didn't even know about the q{} syntax, so I didn't think the stuff I left out would matter, but when using this approach it apparently does. I am going to show the full code here, just so that there are no more suprises.
>
> Sorry for this becoming a "please do my homework" kind of question. I tried for half an hour with no idea how to fix the errors I'm seeing. Something always fails.
>
> The full implementation of OUTPUT_REPRO_CASE looked like this:
>
>
> enum MAKE_FUNCTION_REPRODUCIBLE(alias func, s64 execution_count_to_log_reproduction_on = -1) =
>   "static s64 execution_count = 0;
>   if(context.settings.debug_.output_call_counts_for_reproducible_functions)
>     dlog(\""~fullyQualifiedName!func~" call count: %lld\", execution_count);
>
>   if(execution_count == "~execution_count_to_log_reproduction_on.stringof~"){
>     dlog(generate_reproduction_for_current_function_call!(
>       "~fullyQualifiedName!func~",
>       ["~get_string_argument_list_of_function!([ParameterIdentifierTuple!func])~"])(
>       context.temp_allocator,
>       "~get_argument_list_of_function!([ParameterIdentifierTuple!func])~"));
>
>       trigger_breakpoint();
> }";
>
> It calls the function
>
> String generate_reproduction_for_current_function_call(alias func, string[] parameter_names, Parameter_Types...)(Allocator* allocator, Parameter_Types parameters);
>
> that actually creates the code-string. I did a lot of trial and error getting the trait-calls to expand at the right time (become part of the mixin string or outside of it) and falling on my nose about when something is a type vs a string and so on, so this became really ugly. If you have a better suggestion I'm open to it.
>
> I was testing this inside the function
>
> Dynamic_Array!(Polygon) clip_polygons(Slice!Vector2 p0, Slice!Vector2 p1, bool return_difference_instead_of_union = true, Slice!(Slice!Vector2) p0_holes = Slice!(Slice!Vector2)(), bool debug_print = false);
>
> in which the OUTPUT_REPRO_CASE expands to:
>
> static s64 execution_count = 0;
>   if(context.settings.debug_.output_call_counts_for_reproducible_functions)
>     dlog("vbl.clip_polygons call count: %lld", execution_count);
>
>   if(execution_count == -1L){
>     dlog(generate_reproduction_for_current_function_call!(
>       vbl.clip_polygons,
>       ["p0", "p1", "return_difference_instead_of_union", "p0_holes", "debug_print"])(
>       context.temp_allocator,
>       p0, p1, return_difference_instead_of_union, p0_holes, debug_print));
>
>       trigger_breakpoint();
> }
>
> When I try to change the whole thing to be a token string, execution_count_to_log_reproduction_on was suddenly undefined. I changed this to be an abominational mix of token string and string literal and then it sort of worked, except then the whole business of making the parameter names expand as actual tokens and not strings failed for some reason.
>
> I am super confused (this is a reccuring theme with D compile time stuff on my part :/). Can you show me how this can work with your token string solution?

Maybe you need another mixin in the mixin. The first mixin gets the symbol to the current function. Information from the symbol is used to generate the second mixin. It could be done like this:

import std.traits, std.stdio;
string generateLogCode(alias func)()
{
    string code = "writeln(";
    code ~= "\"" ~ fullyQualifiedName!func ~ "(\"";
    foreach(i, param; ParameterIdentifierTuple!func)
    {
        if(i)
            code ~= ", \", \"";
        code ~= ", " ~ param;
    }
    code ~= ", \");\");";
    return code;
}
enum OUTPUT_REPRO_CASE = q{
    pragma(msg, generateLogCode!(__traits(parent, {}))());
    mixin(generateLogCode!(__traits(parent, {}))());
};
void test(int a, int[] b)
{
    mixin(OUTPUT_REPRO_CASE);
}
void main()
{
    test(1, [2, 3]);
}

On Saturday, 23 October 2021 at 20:24:32 UTC, Tim wrote:
> import std.traits, std.stdio;
> string generateLogCode(alias func)()
> {
>     string code = "writeln(";
>     code ~= "\"" ~ fullyQualifiedName!func ~ "(\"";
>     foreach(i, param; ParameterIdentifierTuple!func)
>     {
>         if(i)
>             code ~= ", \", \"";
>         code ~= ", " ~ param;
>     }
>     code ~= ", \");\");";
>     return code;
> }
> enum OUTPUT_REPRO_CASE = q{
>     pragma(msg, generateLogCode!(__traits(parent, {}))());
>     mixin(generateLogCode!(__traits(parent, {}))());
> };
> void test(int a, int[] b)
> {
>     mixin(OUTPUT_REPRO_CASE);
> }
> void main()
> {
>     test(1, [2, 3]);
> }

That worked! I needed to modify it a bit, since there is the "execution_count_to_log_reproduction_on"-template parameterto the enum. If I try to use that inside the token string, it just says it can't find the identifier, so I had to inject some non-token string in the middle that plays nice with the token string, where I can actually reference the parameter.

For completeness, here is how the code looks like now (I am completely horrified by the result):

> string generate_call_to_log_reproduction(alias func)(){
>   string code = "dlog(generate_reproduction_for_current_function_call!("~fullyQualifiedName!func~", [get_string_argument_list_of_function!([ParameterIdentifierTuple!(__traits(parent, {}))])])(context.temp_allocator";
> 
>   foreach(i, param; ParameterIdentifierTuple!func)
>     code ~= "," ~ param;
> 
>   code ~= "));";
>   return code;
> }
> 
> string generate_call_to_log_execution_count(alias func)(){
>   return "dlog(\""~fullyQualifiedName!(func)~" execution count: %lld\", execution_count);";
> }
> 
> enum OUTPUT_REPRO_CASE(s64 execution_count_to_log_reproduction_on = -1) = q{
>   static s64 execution_count = 0;
>   if(context.settings.debug_.output_call_counts_for_reproducible_functions){
>     mixin(generate_call_to_log_execution_count!(__traits(parent, {})));
>   }
> } ~ "if(execution_count == "~execution_count_to_log_reproduction_on.stringof~")"~q{
>   {
>     mixin(generate_call_to_log_reproduction!(__traits(parent, {}))());
>     trigger_breakpoint();
>   }
> 
>   execution_count++;
> };


and the usage just looks like this:

> void some_function(int a, int b){
>   mixin(OUTPUT_REPRO_CASE!());
>   return a + b;
> }

The "mixin(OUTPUT_REPRO_CASE!());" then expands to:

> static s64 execution_count = 0;
> if(context.settings.debug_.output_call_counts_for_reproducible_functions){
>   mixin(generate_call_to_log_execution_count!(__traits(parent, {})));
> }
> 
> if(execution_count == -1L){
>   mixin(generate_call_to_log_reproduction!(__traits(parent, {}))());
>   trigger_breakpoint();
> }
> 
> execution_count++;


Thanks for your help!!!

On Sunday, 24 October 2021 at 14:05:35 UTC, Simon wrote:
> On Saturday, 23 October 2021 at 20:24:32 UTC, Tim wrote:
>> [...]
>
> That worked! I needed to modify it a bit, since there is the "execution_count_to_log_reproduction_on"-template parameterto the enum. If I try to use that inside the token string, it just says it can't find the identifier, so I had to inject some non-token string in the middle that plays nice with the token string, where I can actually reference the parameter.
>
> For completeness, here is how the code looks like now (I am completely horrified by the result):
>
>> [...]
>
>
> and the usage just looks like this:
>
>> [...]
>
> The "mixin(OUTPUT_REPRO_CASE!());" then expands to:
>
>> [...]
>
>
> Thanks for your help!!!

Maybe core.reflect and codegen could help with this when it gets released