Compiler now generates native SSE register math!
This release is all about documentation, but includes a few bug fixes. The C code has much more comprehensive comments now, and even the assembly code in the runtime library gained a few too.
Version 1.3 of MinimalBASIC compiler released
Version 1.2 of MinimalBASIC compiler released
Fix self-test code for symbol_table. Omit generating code for READ, INPUT, SQR
support when not needed. Improve FOR loop code generation, converting it to
use macros. Fix bugs in symbol table where DATA statement numeric constants
were stored as floats instead of doubles when not using -s. Improve array
bounds checking to work with only one test per index. Enforce a maximum array
size of 10000 instead of just failing when a larger dimension is specified.
I am happy to announce that I finally finished the lua version of the BASICK interpreter and have released BASICK-1.0. This project contains a simple integer BASIC interpreter (no strings, floats, files, or graphics) that has a syntax similar to old-style ROM BASIC implementations. What is special is that you get 4 versions of the interpreter. These are written in Python3.3.x, PHP5.5.x, ruby2.x, and lua5.2.x. The lua version was very difficult to debug and unpleasant to work on, but at last it is done.
I am happy to announce that version 1.0 of the ECMA-55 MinimalBASIC compiler has been released.
In MinimalBASIC-20140408.tar.xz, the peephole optimizer now removes superfluous sequences of 'pushsaddr' immediately followed by 'popsaddr %rdi'.
Optional wide 132 column output mode now implemented. In the process, a few more small bugs showed up and are fixed. Wide mode shows 16 significant digits in float display instead of just 7, and uses 3 digit exponents instead of 2. A cosmetic typo in NBS test 27 is fixed. A cosmetic bug in NBS test 203 that only shows up with wide mode is fixed. Assembly code for print subsystem has been updated to use defined constants instead of magic numbers, and have better stack usage.
MinimalBASIC-20140402.tar.xz is a release candidate for the 1.0 release. All known bugs are again fixed, batch INPUT works, the test harness has been improved to show percentages of PASS/FAIL/Unknown and reasons are shown when a test does not PASS. A simple standalone peephole optimizer that works on the generated assembly code has been added to remove the back-to-back 'pushxmm0 '/'popxmm 0' sequences that tend to get generated occasionally by the expression handling, and the compile scripts and test harness are updated to use it. 207 of the 208 NBS tests pass using the test harness. Test #131 is designed to generate different output every run since it tests RANDOMIZE, so it cannot be automatically verified. I have manually verified it works.
Now both the 32 and 64 bit modes have identical tests passing, and the coredumps in the 64bit read of unquoted but valid numeric values into string variables is fixed. Batch INPUT is the only (known) problem remaining.
There is now a check target in the Makefile to run the NBS test suite. Thanks to Emmanuel Roche, I now have the entire NBS Minimal BASIC test suite. One glaring problem that the test suite automation made clear was the batch input using file redirection does not work. Also, two files that work fine with 32bit mode coredump in 64bit mode (test 93 and 95), which shows READ/DATA still needs some work. The latest snapshot does include all of this, as well as fixes for 64bit INPUT of numeric values, a bug in SGN() for 32 bit, a failure to notice EOF on INPUT, and more.
Testing revealed a bug in ON..GOTO expression handling code generation (was still using 32bit method even in 64bit mode). Also, the -s option to ecma55 was not working. These problems have been fixed.
The arithmetic operations now all default to 64bit doubles.
All features for Minimal BASIC have been implemented. All the NBS tests I have are passing. I shook out a few remaining bugs in handling of spaces, READ/DATA, and INPUT and no known bugs exist at this time. The emphasis of the project will now switch to more rigorous testing to verify the feature for which I cannot get the NBS tests sources.
The new INPUT subsystem has been written, and after some hard work all KNOWN bugs are gone, and the NBS tests for INPUT that I have all pass. If you have a copy of NBS MinimalBASIC test #109, I'd love a copy - that's the one INPUT test I don't have.
All known numeric formatting problems are fixed. I have entered all the NBS tests that I could find, and many of them pass now. The last remaining major feature is complete support for INPUT according to the ECMA55 standard, and work will begin on that next. Note that there is a bug in EXP() for negative value that I still haven't managed to fix yet. Otherwise, most genuine Minimal BASIC programs that do not use INPUT should work now.... read more
I have switched to using David M. Gay's dtoa and g_fmt routines and this has helped a lot towards getting NBS tests 9 and 10 to pass. I still haven't fixed output for things like .000022 (I get 2.2E-5) but decided to go ahead and commit what I have since it's a big improvement. The code I used before wasn't "wrong" but I just was not able to make it output in the admittedly peculiar way I need for this project.... read more
I learned that the exception flags in the MXCSR are 'sticky' and must be reset if you want to continue after detecting an exception. The binary_add, binary_subtract, binary_multiply, and binary_divide macros have been updated for this. I also learned that 0/0 does NOT give a divide by zero exception, but instead gives an Invalid Exception so I had to update binary_divide for that.
I have fixed the code generation so that a divide by zero or an arithmetic overflow on binary add, subtract, divide, or multiply will print an execption, then substitute the appropriately signed infinity and continue as required by section 8.6 of ECMA-55.