On Tue, 30 Jun 2020, Chris Muller wrote:

> Hi Levente and Jakob,
>
>       > Am Mo., 29. Juni 2020 um 01:43 Uhr schrieb <commits at source.squeak.org>:
>       >>
>       >> + ----- Method: MCDirectoryRepository>>includesVersionNamed: (in category 'versions') -----
>       >> + includesVersionNamed: aString
>       >> +
>       >> +       | comparable |
>       >> +       comparable := ((aString endsWith: '.mcz') and: [ aString size > 4 ])
>       >> +               ifTrue: [ aString allButLast: 4 ]
>       >> +               ifFalse: [ aString ].
>       >
>       > Thank you for the suggestion. Why not use aString asMCVersionName
>       > versionName here?
>
>       Copy-paste from parent.
>       When Chris introduced MCVersionName, he tried to use #asMCVersionName
>       there, but later he reverted it according to the history of the parent
>       method. It probably broke stuff (see below). 
> 
> 
> Hmm...  that's true, unfortunately, my comment only says, "Fix" it with no other details.  I can't help but wonder if it was due to a #class test somewhere..
>  
>
>       I think MCVersionName is something that should not exists in its current
>       form. Why?
>       - it's a subclass of ByteString, which limits the possible character set
> 
> 
> It's common in the IT industry for Name identifiers to be restricted to a subset of ASCII to support interoperation with as many types of systems as possible.  It seems an appropriate choice for MC Version names.
>  
>       - #= is not commutative when one object is an MCVersionName, the other is
>       a ByteString:
>               'foo' = 'foo.mcz' asMCVersionName. "==> false"
>               'foo.mcz' asMCVersionName = 'foo'. "==> true"
>       - the above property breaks the contract of #hash and #=, so they
>       sometimes misbehave in hashed collections
> 
> 
> Not in actual practice, though.  The intent was that all uses of version names would be ensuring they're dealing with the first-class MCVersionName.  If one, theoretically, were to mix Strings and MCVersionNames as in your
> example, yes, it would misbehave so... "don't do that."  :)

If it shouldn't be mixed with strings, why is it a string?


Levente

> 
>  - Chris
> 
>