Duration class>>#readFrom: is SURPRISING [was EndOfStream performance cost]

nicolas cellier ncellier at ifrance.com
Mon Nov 12 22:13:04 UTC 2007

While chasing stream pastEnd readers, I ran all 7141 tests and got only 
340 pastEnd reads, and still no failure (except known ones in this 
image), no image crash, not the smell of a bug yet (I know, test 
coverage is quite thin...).

However, seeing Duration class>>#readFrom: among pastEnd readers, i 
inquired implementation and soon noticed a surprising result: 

Reading pastEnd and testing isNil is absolutely legitimate here, I do 
not propose to change it per se.
But I would like this readFrom: behaviour changed, wouldn't you?

At least hope my EndOfStream inquiry won't be totally useless...


More information about the Squeak-dev mailing list