summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorDavid Young <dyoung@hdfgroup.org>2020-07-15 21:26:52 (GMT)
committerDavid Young <dyoung@hdfgroup.org>2020-07-15 21:26:52 (GMT)
commit767afc60bea892abb1ab2ff56473cb50818c042e (patch)
treeff0b3739a8f99acb8e7f8f940754b75370c87ecd /src
parenta0be3626c803caa2f4e0d9b3a4f5b51fc47e280a (diff)
downloadhdf5-767afc60bea892abb1ab2ff56473cb50818c042e.zip
hdf5-767afc60bea892abb1ab2ff56473cb50818c042e.tar.gz
hdf5-767afc60bea892abb1ab2ff56473cb50818c042e.tar.bz2
Update comment.
Diffstat (limited to 'src')
-rw-r--r--src/H5FDvfd_swmr.c13
1 files changed, 3 insertions, 10 deletions
diff --git a/src/H5FDvfd_swmr.c b/src/H5FDvfd_swmr.c
index fe977a3..a7b6d33 100644
--- a/src/H5FDvfd_swmr.c
+++ b/src/H5FDvfd_swmr.c
@@ -1296,20 +1296,13 @@ H5FD__vfd_swmr_index_deserialize(const H5FD_vfd_swmr_t *file,
HGOTO_DONE(FALSE);
/* If the index magic is incorrect, then assume that is a
- * temporary error such as a "torn write." Try again.
+ * temporary error and try again.
*
* XXX XXX XXX
* Under the new protocol, where the index is written in
* one write(2), and the header is written in a distinct
- * second write(2), it is reasonable to expect that the
- * index-write is complete when the index-read occurs.
- * So we should not read bad magic because we read a
- * "torn" write.
- *
- * (I am not sure I believe any recent version of UNIX or
- * Linux suffers from torn writes on local filesystems!
- * Linux manual pages indicate that there was an issue, but
- * it was fixed.)
+ * second write(2), and the header and index are read in
+ * the reverse order, the index magic usually will be intact.
*
* It is possible under the new protocol that we read
* the header on tick `t`, then an arbitrary delay