VM builds still buggy

Brian T. Rice water at tunes.org
Sun Aug 1 01:11:22 PDT 2004


The discovery and repair were made today and discussed on IRC. Here is a 
hopefully clear explanation:

The newBlock opcode was allocating objects that weren't making it onto 
the VM's RootStack (where GC reference-tracing starts). Basically an 
invariant we have to enforce is that objects referred-to from the C 
stack don't get moved around invisibly, and this was being broken. So 
the garbage collector has been altered to have a conservative sweep 
phase, which avoids this issue.

Lee has performed a full VM-hosted bootstrap from scratch and posted his 
output VM sources and image files to the alpha directory. I downloaded 
them and reproduced his results, so all should be well! Please report 
any problems; I will in the meantime perform some code clean-ups and 
prepare for a final release of 0.3 (whew!).

Lee Salzman wrote:
> There are still miscellaneous crashes happening in the final stages of the VM builds. If anyone finds out why, let me know.
> 
> Lee

-------------- next part --------------
A non-text attachment was scrubbed...
Name: water.vcf
Type: text/x-vcard
Size: 208 bytes
Desc: not available
Url : /archives/slate/attachments/20040801/c17d0b60/water.vcf


More information about the Slate mailing list