Comment #2 on issue 306 by
[hidden email]: handling WideStrings
instances in mcz files
http://code.google.com/p/glassdb/issues/detail?id=306Ensuring this in near to impossible.
This is due to the limitations of the underlying DataStream file-out
mechanism.
Eventually, Monticello’s snapshot.bin is just a serialization of the
MCDefinitions.
Hence, if MCDefinitions contain WideStrings somewhere in them (wich may be
the case,
when some source contains wide strings) they are just written out as is.
There is a special handling of strings in DataStream that was last changed
when ByteStings vs. WideStrings
were introduced in Squeak around 2005.
> While ByteStrings are stored by content and marked as
strings, WideStrings are just stored as binary content and marked as of
class “WideString”
But my point is, we are not to change this. There are MCZ-files out there
that were generated that way and we cannot change it.
IMHO, we shall
• support the WideSting named Strings in the way that we convert them on
the fly while loading.
(this is necessary as we cannot fall back to reading the .st-source, as
it may be in UTF-32 once
_any_ source contains a wide character, and there is no possibility to
know that upfront.)
• Move away from DataStream soon. But this is another, major discussion.
BTW: what about the conversion of WideString to QuadByteString I proposed
earlier?
Locked
