Debug information tends to be better when things are compiled.
When uncompiled, backtraces mention the interpreter code instead of the interpreted code. Uncompiled tests make it harder to see where in the tests things go wrong (when things are more complicated than '(test-equal? (foo "bar") "baz")', or maybe in opposite order).
Autocompilation is compilation. Though best don't do autocompilation, 'make clean' doesn't remove the stuff, can get problematic in case of inlining, changes in syntax ... Also it makes logs noisier.
Best regards, Amélie Devos
