Why not gnat Ada in gcc?

[Robert Dewar]

<<According to the regression tester, since inception about 50% of the
test runs (it's continuous, so a test run happens about every three
hours if the flow of patches is sufficiently constant) were full
passes, that is the result was as good as the best previous result.
We seem to be doing marginally better recently, but more 60% not 90%.
This is for one target, other targets probably do worse.
>>

Right, my 90% is indeed across all targets. For any single target,
the figure for us would be more like 98%.