On 6/29/24 01:00, jr wrote:
> hi,
> 
> just a couple of thoughts.
> 
> William F Pokorny <ano### [at] anonymousorg> wrote:
>> ...
>> One day, not long ago, it hit me that maybe identifier prefixes would be
>> a way we could implement reasonably broad, useful constant-ness - ...
>>
>> Given I'm going to delay to implement a build configuration option for
>> this feature, do you think some other prefix would be better? ('_r',
>> 'ro_', 'r_', ?)
> 
> "democracy" eh ?! </grin>.  given those options, 'ro_' would be my choice, as
> it's "mnemonic".
> 

Yeah, and ro_ fits with my keyword prefix direction too.

> 
>> Aside 1: I have still bugs and issues with the current global / local
>> dictionary behavior in v3.8+ to sort... Our implementation doesn't match
>> our documentation and we probably need a #top (or #upid (upvar n))
>> directive alongside #directive and #local.
> 
> having read ingo's Nim language examples, I really like the "everything is
> immutable unless" approach.  safe.  perhaps all macro args (as you wrote
> somewhere) ought to be "ro" by default, and an explicit 'upvar N' required for
> every exception ?  (that (upvar N) :-) would be _seriously_ nice to have)
> 

I've been thinking about those examples too & my response to Ingo's 
question. I failed / bailed on my initial attempts to make a macro 
parameters constant always - but success would have meant a SDL change 
much more likely to break current code than an 'ro_' prefix. I hadn't 
thought much at all about a keyword to change defaulted read-only to 
something mutable via a new keyword.

It might be we could check that the mutability of the parameter in the 
macro matches the caller's ID mutability, but there are limits to how 
much checking like that we can do given SDL isn't compiled and linked. 
We'd be paying the price of the overhead on every call to the macro.

 From what I've seen people mostly use the macro 'pass by reference and 
change' and '#declare ID' as mechanisms to return multiple values or to
define what they want global as with the yuqk munction concept compiling 
users functions only once. (We now have the tuple return and assignment 
mechanisms in v3.8.)

Another thing which might be possible nearer term, would be to make 
mixed use of #declare and #local with the same variable name inside 
macros illegal in yuqk (or in just the debug compile).

See the attached file for a few examples of how mixing one identifier 
name with both #local and #declare can result in instability. The macro 
parameter - change by reference - issue aside.

We should mention too, for those stumbling across this thread later, 
that a good solution for all the macro-ish exposures to inadvertent 
identifier change is to use functions when dealing with just math and a 
return a value.

Bill P.

Post a reply to this message

Attachments:
Download 'instabilitywithlocalanddeclareuse.txt' (3 KB)