Change Summary:
it did turn out the libzpool is also building lz4, also need to add FAKEKERNEL case.
Commit: |
|
||||
---|---|---|---|---|---|
Diff: |
Revision 2 (+151 -1555) |
Change Summary:
fix libstand build
Commit: |
|
||||
---|---|---|---|---|---|
Diff: |
Revision 3 (+151 -1555) |
Change Summary:
and build fix for libficl.
Commit: |
|
||||
---|---|---|---|---|---|
Diff: |
Revision 4 (+154 -1557) |
-
Changing the LZ4 code used by ZFS should be done with care, because it impacts the on-disk format. Changing the compressed bytesteam at all can have unexpected consequences, even if it can still be decompressed by older ZFS LZ4 code. I think the areas to look out for are l2arc (w/ compressed ARC disabled) and "nopwrite". Allan Jude has investigated this in the context of trying to update the LZ4 code in ZFS to a later version of LZ4.
It isn't impossible to have one LZ4 implementation for ZFS and other uses, but you may find that the other use cases have requirements that are at odds with ZFS's. (e.g. other use cases would prefer to always have the most up-to-date LZ4 code, whereas the considerations mentioned above require that we stick with older LZ4 code for ZFS).
Change Summary:
Missed FAKEKERNEL cases. The one source for lz4 needs to cope with different endian detection APIS - loader is using sys/endian.h.
Commit: |
|
||||
---|---|---|---|---|---|
Diff: |
Revision 5 (+167 -1557) |
-
Can you add links to diffs of the other
lz4.c
versus the new ones, so we can see how they (logically) change too?
Change Summary:
cast align issue, and left over #.
Commit: |
|
||||
---|---|---|---|---|---|
Diff: |
Revision 6 (+167 -1558) |