Doesn't have POSIX robust mutexes - GNU userland on BSD kernel

- DISTRIBUTE_GROUP_DOC makes doxygen give several fields the
  same doc: mn_hi + mn_lo in MDB_node.
- With mdb_mutex_t + mdb_mutexref_t, instead split them up.

Never clear mc_error, we could lose a failure in the other thread.

Lorenz Bauer authored 8 years ago and Hallvard Furuseth committed 8 years ago

Return EPIPE instead.

Other functions depend on them to do so.
For mdb_node_read(), instead remove such a dependence.

No change in functionality.

No change in functionality, except needs less mallocing.

The 0xFFFD check seems due to misleading MultiByteToWideChar() doc.
Bad UTF-8 gives 0xFFFD in the output string, not the return value.

...and check !MDB_CLOEXEC in an 'if' rather than '#if'
to match its non-zero usage.

We can run without F_NOCACHE if it fails. And we do not know
what combining it with O_DIRECT means, if a system has both.

Howard Chu authored 8 years ago and Hallvard Furuseth committed 8 years ago

It usually gets done anyway, but one of the fastpath shortcuts
bypassed this step.

Make explicit and default nonzero equivalent. Parenthesize.

1st struct member was not a scalar.

Preserve DB flags (use metapage#1) when main DB is empty.
Fail if metapage root != actual root in output file.

Handle errors.  Fix cond_wait condition so mc_new
is the sole control var.  Drop specious cond_waits.
Do not look at 'mo' while copythr writes it.

Don't know if posix_memalign() pointer is defined after failure.
Some _aligned_free() doc seems to say arg NULL = user error.

Howard Chu authored 9 years ago and Hallvard Furuseth committed 8 years ago

Logical counterpart to GET_MULTIPLE, NEXT_MULTIPLE

Howard Chu authored 8 years ago and Hallvard Furuseth committed 8 years ago

Revert excess cursor fixup
xcursor fixup depends on init state