[Vm-dev] Underscore for assignment in VMMaker and related plugins

[Eliot Miranda]

Hi Dave, Hi Marcel,

> On Aug 8, 2021, at 5:40 PM, David T. Lewis <lewis at mail.msen.com> wrote:
> 
> 
> Background: By convention the assignment operator in modern day Squeak
> is ':=' and it Cuis it is '_'. The actual fonts are rendered nicely in
> either case for Cuis and Squeak.
> 
> The VMs used for Cuis and Squeak consist of the VM proper along with various
> VM plugins such as OSProcessPlugin and VectorEnginePlugin. The VM plugins
> are written in Smalltalk ("slang") with supporting external code as needed.
> The Smalltalk code for each VM plugin is translated to C code, and the
> resulting C code is expected to be correct regardless of the dialect of
> Squeak/Cuis in which the plugin is written.
> 
> The generated C code for a VM plugin is the same regardless of whether the
> Smalltalk source is written in the '_' or ':=' convention. However, mixing
> these two conventions in the same VMMaker image may be confusing, especially
> if the (Squeak) image has disabled its 'Allow underscore selectors' preference.
> 
> OK, enough background. The question I want to present is this:
> 
> The new VectorEnginePlugin is being developed in Cuis, and therefore uses
> the '_' convention in its Smalltalk source code. When building a new
> opensmalltalk-vm VMMaker image from primary sources, the 'Allow underscore
> selectors' preference' is disabled, presumably in order to catch VM plugin
> code that accidentally uses the '_' convention. However, in the case of
> VectorEnginePlugin, the use of '_' is intentional, and we can expect it
> to be used for current and future versions of the plugin Smalltalk code
> as development proceeds.
> 
> So should we:
> 
>  1) Relax the restriction and allow VectorEnginePlugin to be loaded without
>     modification in a VMMaker image
> 
> or:
> 
>  2) Convert current and future versions of the VectorEnginePlugin Smalltalk
>    to use ':=' for assigments prior to loading in a Squeak VMMaker image
>    (method author/timestamps would be preserved).
> 
> I personally prefer 1) but consider either option to be reasonable and doable.
> 
> Comments? Preferences?

My preference is 1a :-). 1a is supporting a Compiler patch that allows one to assert the 'Allow underscore selectors' preference on a per-class basis and adding some class side state (a method? A method containing a pragma?) to VectorEnginePlugin that asserts the preference.  Maybe this is too elaborate, and there’s a bootstrap problem (the class side method must be loaded before methods containing a underscore assignments) but I like not having the preference globally.

> Dave