Network

 
AuthorCommitMessageCommit Date
David YoungDavid Young
295295f957dRemove disused #ifdef QAK code.
David YoungDavid Young
21e2c55c5a8Be brief: don't cast the last parameter of H5Pset_file_space_strategy(), the constant `1`, to `hsize_t`.
David YoungDavid Young
a56d19800ccDon't see a 1GB threshold for tracking free filespace, that workaround for VFD SWMR was never 100% effective, and now that there is a delay line on filespace frees, it is not necesesary.
David YoungDavid Young
ac286ab26bfMerge all of my changes from merge-back-to-feature-vfd_swmr-attempt-1, including the merge of `hdffv/hdf5/develop`, back to the branch that Vailin and I share. Now I need to put this branch on a fork with a less confusing name than vchoi_fork!
Vailin ChoiVailin Choi
9fb937b73d3Fix the FSM bug when setting the FSM threshold to a non-default value. Check for smaller or larger section size after merging and shrinking a section, for this case is the section that is smaller than threshold (see H5MF_xfree() in H5MF.c). It is possible for the section to be smaller after merging/shrinking (see H5MF__sect_large_shrink() in H5MFsection.c).
David YoungDavid Young
f4506053379Straggler from last commit: add test/vfd_swmr_common.c.
David YoungDavid Young
901ca275a12Make all VFD SWMR test writers wait for a signal before closing their HDF5 file, and always send that signal after the readers are finished. Carve out an exception, though, for the expand/shrink test, where vfd_swmr_writer is used to initialize the file. Move await_signal() to vfd_swmr_common.c for sharing with all of the writers. Add to vfd_swmr_writer a command-line option flag, -W, that stops it from waiting for a signal before closing the HDF5 file and quitting. While I'm here, rewrit...
David YoungDavid Young
60c49e6f3ebRemove a calloc cast that was just for C++ compatibility and really uglied-up the code.
David YoungDavid Young
c396d37945dDefer filespace frees performed by VFD SWMR writers so that raw-data space is not reused prematurely. Probably should optimize this by freeing only the freeing of raw-data regions is deferred.
David YoungDavid Young
19577f797e8I wish that I had known previously that developers were cheating the cross-platform property-list test, because then I could have cheated, too, and saved myself a lot of time. Regenerate and then copy the 64-bit, little-endian version of the file over all other-bits and other-endian versions. I've filed a ticket about this.
David YoungDavid Young
0dbc26bec92Use a for-loop instead of a while-loop to walk another linked list, taking care of single-entry deletions. Will use LIST_FOREACH_SAFE() or a similar queue(3) macro, eventually.
David YoungDavid Young
def41b41944Use a for-loop instead of a while-loop to iterate a linked list in the face of possible single-element deletions. Eventually I will use LIST_FOREACH_SAFE() or some similar queue(3) macro.
David YoungDavid Young
c2514497530Use a for-loop instead of a while-loop to search a linked list, and perform the operation on a found item (if any) *outside* of the loop. Eventually this for-loop will use a macro from queue(3).
David YoungDavid Young
73076b3999eRewrite a while-loop that walks a linked list as a for-loop. Eventually I will replace the for-loop with a macro like LIST_FOREACH() from queue(3).
David YoungDavid Young
80ea5c46218Use a for-loop to walk a linked list. Eventually this will use a macro such as LIST_FOREACH() from queue(3).
David YoungDavid Young
f5d68474fa3Delete some excess whitespace and leave a comment to myself about why it's safe to remove a page-table entry for freed file-space storage. NFCI.
David YoungDavid Young
8746fb92554Rename a variable, s/pbe_ptr/entry/, part 2. While I'm here, use a for-loop instead of a while-loop to iterate over a linked-list of entries. Eventually I will turn the for-loop to a use of LIST_FOREACH() or a similar macro from queue(3).
David YoungDavid Young
74cfb6cda2dSimplify a variable name, step 1. s/pbe_ptr/entry/ in some comments.
David YoungDavid Young
8f0fd4f1073Reduce parenthesization. NFCI.
David YoungDavid Young
bc936b2522aWe never read through the metadata accumulator, so don't mention it in an error message.
David YoungDavid Young
8fbdb68133dUse the right format strings for uint32_t, uint64_t.
David YoungDavid Young
c5e5b2f0b86Make a copy of NetBSD's <sys/queue.h> in src/bsdqueue.h.
David YoungDavid Young
03205d1b9e9Use struct assignment instead of memcpy.
David YoungDavid Young
023e69990d8Fix typo in an error message.
David YoungDavid Young
5ec4a554b27Delete gratuitous casts.
David YoungDavid Young
21c3cb07229Use proper format strings in some diagnostic printfs. NFCI.
David YoungDavid Young
6efcb7d1d7dAssert that a raw-data page we are examining *does* intersect the interval [addr, addr + size) that we are trying to read, replacing an assertion that did not make any sense in context. I'm not sure why it took so long for the nonsensical assertion to fire, but maybe it was because we didn't do any multi-page raw-data reads through the page buffer, before?
David YoungDavid Young
03eebd31186Reduce code duplication in enc_dec_plist_cross_platform.c: extract a subroutine that computes the name for a plist file, reads and decodes it.
David YoungDavid Young
51250a4adb3Extract a subroutine, vfd_swmr_pageno_to_mdf_idx_entry(), that performs the binary search for the shadow-file entry corresponding to a page number. Rewrite to avoid awkward casts between unsigned and signed integers. Use the new routine in a couple of places.
David YoungDavid Young
e1e48314fe7Mention the delay on freeing file space for VFD SWMR.
David YoungDavid Young
55f52917ccdAs an experiment, let's serialize property lists into a zeroed buffer.
David YoungDavid Young
8eebac52728Remove #if 0'd quick fix for the free-space management bug in src/H5MF.c. Also, squashed commit of the following: commit adcf8a315e82c0848d126e7e46b662930c081896 Author: Vailin Choi <vchoi@jam.ad.hdfgroup.org> Date: Mon Dec 2 11:07:25 2019 -0600 Fix the FSM bug when setting the FSM threshold to a non-default value. Check for smaller or larger section size after merging and shrinking a section, for this case is the section that is smaller than threshold (see H5MF_xfree() in H5...
David YoungDavid Young
3b70a15c3deRegenerate 64-bit, little-endian test property lists.
David YoungDavid Young
6181697bc04Regenerate 32-bit little-endian test property lists.
David YoungDavid Young
74dfbdecc85Regenerate the 64-bit, big-endian property list test files.
David YoungDavid Young
bdbf7b79481Regenerate plists for big-endian.
David YoungDavid Young
3f5fb5fec46If a compilation unit #include a header that defines a static inline function, and that static inline function calls an external function, then even if you do not use the inline function, the Solaris Studio compiler produces a reference to that external. That behavior prevented H5make_libsettings from linking: H5_nanosleep, used by the h5_retry_ functions, was not found. #including H5retry_private.h only where it's needed fixes the issue.
David YoungDavid Young
a5ff057b1cbQuiet some warnings about change of sign changing values, signed v. unsigned comparisons.
David YoungDavid Young
b8edead516eMMerge remote-tracking branch 'origin/feature/vfd_swmr' into merge-back-to-feature-vfd_swmr-attempt-1
David YoungDavid Young
75aa099cab7MMerge remote-tracking branch 'origin/feature/vfd_swmr' into merge-back-to-feature-vfd_swmr-attempt-1
David YoungDavid Young
847cec21b44We have to use H5VLobject() instead of H5Iobject(). Welcome to the New VOL Order. :-/
David YoungDavid Young
291c28bec70Straggler from last: adjust to the New H5F_shared_t Order, change some signed integers back to unsigned.
David YoungDavid Young
1174ef01bd4MMerge remote-tracking branch 'origin/feature/vfd_swmr' into merge-back-to-feature-vfd_swmr-attempt-1
David YoungDavid Young
4182a7126dbCommit diagnostic changes to free-space management and my work in progress on the VFD SWMR test-file generator so that Vailin can take a look at the bug I found.
David YoungDavid Young
cbe7901ddaeIn the VFD SWMR expand/shrink test, have the reader trail behind the writer by the maximum size of a change in dataset length so that it cannot jump in and read a dataset *after* it's been extended but *before* the data is written. With other changes (forthcoming), this prevents the expand/shrink test from reading unexpected data values.
David YoungDavid Young
346222a8760Do not cork and uncork the metadata cache, that should not be necessary with VFD SWMR.
David YoungDavid Young
15f1cf64c81Quiet the sparse writer: bracket the expected-to-fail H5Aexists_by_name() call by H5Eget_auto/set_auto calls that temporarily disable the error-stack printing.
David YoungDavid Young
45d0761a3e9Add a couple of debug echo(1)s and improve a comment.
David YoungDavid Young
1e1a9c87636Here is my work in progress on debugging the sparse reader/writer test.
David YoungDavid Young
35dc70f1f1eUse h5_retry_init/_try to simplify this code a bit and shorten the staircase.