On Mon, 19 Feb 2018, Martok via Lazarus wrote:

Am 19.02.2018 um 11:10 schrieb Sven Barth via Lazarus:
As long as the code does not rely on undocumented behavior, yes.
And therein lies the issue. Things that worked the same way for >25 years and
are mentioned explicitly in numerous secondary literature are considered

I will mention 1 case in point. Const parameters.

For years, people mistakenly assumed that Const parameters were be passed by reference if they were "big" (records), a behaviour which
Delphi exhibited. And 'mentioned explicitly in numberous secundary
literature' as you put it so nicely. But not documented.

FPC introduced Constref to guarantee a constant was passed by
reference (I think IUnknown methods were the first to use this), but otherwise refused to change our view of 'Const'.

Then one day, someone on the Delphi team (I believe Allan Bauer?) makes a
blog post stating that "Const never implied that big types are passed by
reference", and that people should correct their code and use the new [Ref]
attribute if they want a const parameter passed by reference.

Thus supporting the point of view which we had all along.

So, no, we will not promise to conform to undocumented behaviour.

And documented behaviour is what the language guide of Delphi states.

So unless Delphi _documents_ that


is 100% equivalent to


It remains an implementation detail when the temp is finalized.

Which, in FPC, is currently at the end of the procedure.

Lazarus mailing list

Reply via email to