From: Michael Hope <michaelh@ju...> - 2001-06-05 04:00:43
Evening all. I've commited the latest set of test cases. It perhaps
covers 30% of the basic operations. I've also come to the conclusion that
I really don't understand va_args well enough to write a test case for
them :) The current output from the suite is:
Summary for 'host': 0 failues, 938 tests, 178 test cases
Summary for 'z80': 111 failues, 944 tests, 178 test cases
Note that most of the cases are generated through permutation so there's
really only ~17 test cases and ~93 test points. I don't know where the
extra tests came from.
The next steps are
* Integrate the tests in sdcc/src/regression
* Improve coverage to ~50%
* Add tests cases that cover previous bugs
* Add emulators for the other ports, with mcs51 probably first
* Get other people using it
* Start killing bugs
I haven't numbered these as I don't know what order to do them in. Any
I'm especially interested in test cases covering previous bugs to make
sure that they don't pop up again. Current candidates are structure
member not being invalidated over function calls, and array dereferences
by array dereferences i.e. a[a] but I'd really like others as well.
Also, how can we decide if something is a bug or not? What level of
standard compliance do we want? In the 'implementation dependent' parts,
do we want to pick one and make each port consistent? Should we pick
another compiler to be consistent with, say gcc? I'm thinking of things
like mod of negative numbers - at least the static and runtime forms
should be the same.
I've also taken the liberty of making all new bug reports submitted via
sourceforge come to this list so that we can trak it better.
Get latest updates about Open Source Projects, Conferences and News.