Create the new repo file O_EXCL to prevent racing with another razor process.
And remember to clean it up on exit paths.
1 Towards replacing rpm + yum (0.1):
5 - pre install check; check that dirs can be created (no files where
6 want to create dirs), move config files according to file
9 - run hooks; probably fork to shell and in the parent just do a
10 write() of the hook section to a pipe to the shell.
12 - store rpm headers for installed packages.
14 - rpm seems to consider glibc > 2.6.90 to mean greater than
15 2.6.90-anything. That is, a comparison that doesn't mention the
16 release field, shouldn't regard the release field of pkgs it
17 compares against. glibc-common-2.6.90 has
19 conflicts: glibc < 2.6.90, glibc > 2.6.90
21 since rpm doesn't let you do glibc != 2.6.90, and
23 requires: glibc = 2.6.90
25 will always pull in glibc. But even with a != relation, would
26 glibc-2.6.90-16 be equal to 2.6.90? glibc 2.7.90-8 dropped it in
27 favor of requires = 2.7.90-8 (#225806).
33 - figure out how to canonically represent empty string... ~0?
35 - space calculation before transaction, but ideally, do a number of
38 - pre-link changing binaries and libs on disk screwing up checksum?
40 - nail down byte-order of repo file.
42 - version the sections in the file, put the element size in the header
43 so we can add stuff to elements in a backwards compatible way.
44 maybe not necessary, we can just add sections that augment the
45 sections we want to add to (similar to how rpm has add versioned
48 - pipelined download and install; topo-sort packages in update set,
49 pick one with all deps in the current set, add it to the current set
50 and satisfy deps against update set => result: minimal update
51 transaction. Queue download and install/update transaction for the
52 packages in the minimal set, start over. This also makes the
53 installation phase much more interruptible, basically just stop
54 after a sub-transaction finishes. As we keep the update set around
55 as a target, we can restart if needed. Probably don't need to, can
56 just do a new update. During a sub-transaction we should keep the
57 target set (i.e. the current set to be) around as a lock file
58 (system.repo.lock or so, see git) so that razor updates are
59 prevented if the systems crashes during an update.
63 - keep history of installed packages/journal of package transaction,
64 so we can roll back to yesterday, or see what got installed in the
67 - transactions, proper recovery, make sure we don't poop our package
68 database (no more rm /var/lib/rpm/__cache*).
70 - use hash table for package and property lists so we only store
71 unique lists (like for string pool).
73 - use existing, running system as repo; eg
75 razor update razor://other-box.local evince
77 to pull eg the latest evince and dependencies from another box. We
78 should be able to regenerate a rzr pkg from the system so we can
79 reuse the signature from the originating repo.
81 - Ok, maybe the fastest package set merge method in the end is to use
82 the razor_importer, but use a hash table for the properties. This
83 way we can assign them unique IDs immediately (like tokenizing
86 - test suite should be easy, just keep .repo files around and test
87 different type of upgrades that way (obsoletes, conflicts, file
88 conflicts, file/dir problems etc). Or maybe just keep a simple file
89 format ad use a custom importer to create the .repo files.
93 do { ... } while (((e++)->name & RAZOR_ENTRY_LAST) == 0);
95 idiom for iteration of directories.
97 - overlay package sets? mount a read-only /usr over nfs or from the
98 virt-host and have a local package set overlaid over the read-only
99 one. shouldn't need new features in the core package set data
100 structure, but should be just conventions on top. we have the base
101 package set from the r/o system, the overlay set from the local
102 system and we can have an effective package set which is the merge
103 of everything from the overlay into the base set. the effective set
104 is easy to compute and we could do it on the fly or cache it. or
105 maybe the effective set is the on-disk representation and the
106 overlay can be computed when needed, we just keep a link back to the
109 - incremental rawhide repo updates? instead of downloading 10MB zipped
110 repo every time, download a diff repo? Should be pretty small,
111 especially if we don't have file checksums in metadata. Filenames
112 and properties are for the most part already present, typically just
113 a version bump plus maybe tweaking a couple requires. The upstream
114 repo can store multiple incremental updates in one big file and
115 provide an index file that maps updates for a given date (we should
116 use repo-file checksums though) to a range in the file: Download the
117 index file, search for a match for your latest rawhide.repo file,
118 download range of updates that brings it up to date.
120 - use hash tables for dirs when importing files to avoid qsorting all
125 - eliminate duplicate entries in package property lists.