[Vm-dev] bugs in DeflatePlugin?

Fabio Niephaus lists at fniephaus.com
Mon Jan 10 08:41:45 UTC 2022

Hi Florin,

Happy new year and thanks for the bug report! PNGReadWriterTest seems
to pass on TruffleSqueak (although I remember some bug fixing a while
ago). Anyway, I looked into #primitiveZipSendBlock which seems to be
the only sender of #loadZipEncoderFrom:.

While #loadZipEncoderFrom: reads from writeStreamInstSize + 1 and + 2,
which is off by one as you mentioned, #primitiveZipSendBlock stores at
readStreamInstSize + 1 and + 2. And as it happens, readStreamInstSize
seems to be 4 while writeStreamInstSize is 5, which compensates for
the off-by-one error. However, I don't understand why reading from the
wrong positions does not cause the tests to fail. Anyway, I cleaned
this up in TruffleSqueak (see [1]).


[1] https://github.com/hpi-swa/trufflesqueak/commit/5b6066b54e03e19ec4c4d2b22550ff081e46f6e6

On Mon, Jan 10, 2022 at 4:33 AM Florin Mateoc <florin.mateoc at gmail.com> wrote:
> I should have also mentioned that PNGReadWriterTest also has 18 failures for SqueakJS, and I think I remember them failing in TruffleSqueak as well
> On Sun, Jan 9, 2022 at 10:18 PM Florin Mateoc <florin.mateoc at gmail.com> wrote:
>> Hi all, and Happy New Year!
>> I think I just found a couple of bugs in the DeflatePlugin. They were causing errors in my JsSqueak, I don't fully understand why they don't seem to cause problems in Squeak. Anyway, after I fixed them, the primitives work and the PNGReadWriterTest tests are all succeeding for me.
>> One is in #loadZipEncoderFrom:
>> The method checks and then determines the writeStreamInstSize and then fetches the bitBuffer and bitPosition (the first two instvars of ZipEncoder, a direct subclass of WriteStream) from offsets writeStreamInstSize + 1 and writeStreamInstSize + 2, although the fetches are 0-based
>> The other is in #primitiveZipSendBlock
>> The method, still working on a ZipEncoder, at the end stores the modified values for bitBuffer and bitPosition at offset readStreamInstSize + 1 and readStreamInstSize + 2, which is just wrong.
>> As I said, I don't quite get why this does not cause at least image-side errors if not VM crashes in Squeak, but they should probably be fixed anyway :)
>> All the best,
>> Florin

More information about the Vm-dev mailing list