Please release 1.8.5
wdroste opened this issue · comments
This is critical for us too. Any attempts to use pure-java on a snappy-compressed file end up with a java.lang.IllegalArgumentException: newLimit > capacity
being thrown due to the uncompressed
buffer being resized beyond its capacity due to the wrong use of uncompressed.position()
fixed in #293.
Please release 1.1.8.5 as soon as possible.
@xerial can you please update this. It's been since Oct 20th this fix has been in.
we are also facing this issue, any plans to release this bug fix.
Hello @xerial , thanks you for this library, any update on this one please?
I didnt find any workaround to make it work with Spark 3.2.1 on kubernetes.
Thanks you very much!
@xerial Hi,thank you for this library.
This bug has been fixed for a year.
Would be grateful if 1.1.8.5 could be released.
Hi,
Do you know when the version with this fix will be released?
Regards.
#293 introduces a binary change, and it will not work in some platforms like AIX/ppc,ppc64, s390x, etc., which are not covered by docker-based cross-compilers.
If @odaira can rebuild native libraries for these platforms as he did 4 years ago, it's safe to release. Otherwise, we need to drop support for these platforms.
Pure-java support was an attempt to remove such a dependency on human labor, but it seems the current pure-java snappy code is not properly implemented and caused another issue, so it's blocking the project to ship a new release.
@xerial The change in #293 is meant to affect the pure-java implementation only, and it simply refactors the moment in which the size of the compressed/uncompressed buffers is computed, so that the offset used as argument at the limit operation is not wrongly affected by the previous compression/uncompression operation.
Could you please elaborate on how this fix affects the AIX/ppc, ppc64, etc. environments? Do you mean that, by applying it, only the pure-java implementation would be fixed but not the native libraries? And if so... wouldn't it be possible to release 1.1.8.5 with this fix only, and keep a ticket open for the platforms that cannot benefit from this change?
This library is used by the Apache Hadoop libraries for snappy compression/decompression of large data files in data lakes, which means a really huge amount of corporate users. But in practice this issue makes it unusable in production, forcing anyone who needs to use this in a stable environment to keep their own fork 😞
Released https://github.com/xerial/snappy-java/releases/tag/v1.1.9.0, which no longer supports pure-java mode. Some native libraries for special platforms like AIX are not included.