Other phrasings considered:
- "tractably automate arbitrary manual tests"
- "make it possible to automate arbitrary manual tests"
Thanks Paul Biggar for the feedback.
Now all known discrepancies between C++ and SubX translators are fixed.
If a SubX program builds with C++, it should also build fine with just
SubX.
(If it doesn't build with C++, all bets are off. The self-hosted SubX
translator has negligible error-detection or handling.)
This makes the C++ translator more consistent with the self-hosted
translator.
We go through some contortions to continue supporting unit tests without
'Entry' labels. But we still want to throw good errors when translating
.subx files at the commandline.
A little more resizing of buffers. apps/hex.subx is now building an
identical binary.
I'm now aborting on allocation failures. That requires disabling
a couple of tests. (I'm not quite confident enough of this decision to
delete them outright.) I want to treat all segfaults as bugs, and
machine code is no place to add boilerplate checks for return values of
standard library functions.
Snapshot while debugging survey.subx by print.
I can see the error in 1 minute with this command:
subx run apps/survey < a.pack
(where a.pack is obtained from `ntranslate 049*.subx 05[0-8]*.subx`)
By contrast, using the trace requires 4.5 minutes:
subx --trace run apps/survey < a.pack
It generates a trace of 4.4GB with almost 83M lines.
The trace takes 2 minutes to load.. oops, I forgot to load labels with
`--debug`.
Even though the standard library is building and passing tests, the
binaries it generates aren't exactly bit for bit identical with the
originals. Comparing using `diff_ntranslate`, it looks like the data
segment starting address isn't computed right in survey.subx
(`compute-addresses`) when I start translating layer 058. Deleting some
tests brings the code segment to a p_offset where bits 8-11 (the lowest
4 bits excluding the lowermost byte) are cleared and everything works.
However, if bits 8-11 are set, then they don't make it to p_vaddr and
p_paddr.
Tried reproducing with a unit test, but the unit test passes fine.
Translates 5k lines of input in 26 seconds.
I'm not sure why I need to grow the label table. It was already 512 entries
long, and I'm only using 373 so far.
We can now translate layers 49-72 using the self-hosted translator.
The translator has now demonstrated translation over 4k lines. Most verbose
phase output is 325KB, even if the final binary is 15KB.
Emulation is too slow now, so I'm back to debug by print on a Linux machine.
We can now translate layers 49-56 using the self-hosted translator
(`translate` and `ntranslate`).
As a follow-up to commit 5404, the self-hosted translator is a little
more strict than the C++ translator in 3 places:
a) All .subx files must define a data segment.
b) All .subx files must define an `Entry` label.
c) All numbers must be in *lowercase* hex.
In all cases, where programs work with the C++ translator but violate
the self-hosted translator's assumptions, we must make sure we raise
errors rather than silently emit bad code.
Kartik Agaram