Commit Graph

21 Commits

Author SHA1 Message Date
Ævar Arnfjörð Bjarmason
22d2f70e85 reftable tests: avoid "int" overflow, use "uint64_t"
Change code added in 1ae2b8cda8 (reftable: add merged table view,
2021-10-07) to consistently use the "uint64_t" type. These "min" and
"max" variables get passed in the body of this function to a function
whose prototype is:

    [...] reftable_writer_set_limits([...], uint64_t min, uint64_t max

This avoids the following warning on SunCC 12.5 on
gcc211.fsffrance.org:

    "reftable/merged_test.c", line 27: warning: initializer does not fit or is out of range: 0xffffffff

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-01-13 13:39:09 -08:00
Han-Wen Nienhuys
f2b255141b reftable: avoid initializing structs from structs
Apparently, the IBM xlc compiler doesn't like this.

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-01-13 13:36:34 -08:00
Han-Wen Nienhuys
cd1799dea0 reftable: support preset file mode for writing
Create files with mode 0666, so umask works as intended. Provides an override,
which is useful to support shared repos (test t1301-shared-repo.sh).

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-12-23 12:28:36 -08:00
Han-Wen Nienhuys
0dd44584ab reftable: signal overflow
reflog entries have unbounded size. In theory, each log ('g') block in reftable
can have an arbitrary size, so the format allows for arbitrarily sized reflog
messages. However, in the implementation, we are not scaling the log blocks up
with the message, and writing a large message fails.

This triggers a failure for reftable in t7006-pager.sh.

Until this is fixed more structurally, report an error from within the reftable
library for easier debugging.

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-12-23 12:28:34 -08:00
Han-Wen Nienhuys
019bd34082 reftable: fix typo in header
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-12-23 12:28:28 -08:00
Han-Wen Nienhuys
e793168364 reftable: add dump utility
provide a command-line utility for inspecting individual tables, and
inspecting a complete ref database

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Helped-by: Carlo Marcelo Arenas Belón <carenas@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
e48d427268 reftable: implement stack, a mutable database of reftable files.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
acb533440f reftable: implement refname validation
The packed/loose format has restrictions on refnames: a and a/b cannot
coexist. This limitation does not apply to reftable per se, but must be
maintained for interoperability. This code adds validation routines to
abort transactions that are trying to add invalid names.

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
1ae2b8cda8 reftable: add merged table view
This adds an abstract, read-only interface to the ref database.

This primitive is used to construct the read view of the ref database
(the read view is constructed by merging several *.ref files). It also
provides the mechanism to provide a unified view of the refs in the main
repository and the per-worktree refs.

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
3b34f636df reftable: add a heap-based priority queue for reftable records
This is needed to create a merged view multiple reftables

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
ffc97f1a9e reftable: reftable file level tests
With support for reading and writing files in place, we can construct files (in
memory) and attempt to read them back.

Because some sections of the format are optional (eg. indices, log entries), we
have to exercise this code using multiple sizes of input data

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
46bc0e731a reftable: read reftable files
This supports reading a single reftable file.

The commit introduces an abstract iterator type, which captures the usecases
both of reading individual refs, and iterating over a segment of the ref
namespace.

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
17df8dbeba reftable: generic interface to tables
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
f14bd71934 reftable: write reftable files
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
35425d1034 reftable: a generic binary tree implementation
The reftable format includes support for an (OID => ref) map. This map can speed
up visibility and reachability checks. In particular, various operations along
the fetch/push path within Gerrit have ben sped up by using this structure.

The map is constructed with help of a binary tree. Object IDs are hashes, so
they are uniformly distributed. Hence, the tree does not attempt forced
rebalancing.

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
e581fd7231 reftable: reading/writing blocks
The reftable format is structured as a sequence of block. Within a block,
records are prefix compressed, with an index of offsets for fully expand keys to
enable binary search within blocks.

This commit provides the logic to read and write these blocks.

Helped-by: Carlo Marcelo Arenas Belón <carenas@gmail.com>
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
e303bf22f9 reftable: (de)serialization for the polymorphic record type.
The reftable format is structured as a sequence of blocks, and each block
contains a sequence of prefix-compressed key-value records. There are 4 types of
records, and they have similarities in how they must be handled. This is
achieved by introducing a polymorphic 'record' type that encapsulates ref, log,
index and object records.

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
1214aa841b reftable: add blocksource, an abstraction for random access reads
The reftable format is usually used with files for storage. However, we abstract
away this using the blocksource data structure. This has two advantages:

* log blocks are zlib compressed, and handling them is simplified if we can
  discard byte segments from within the block layer.

* for unittests, it is useful to read and write in-memory. The blocksource
  allows us to abstract the data away from on-disk files.

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
ef8a6c6268 reftable: utility functions
This commit provides basic utility classes for the reftable library.

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Helped-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
890044708d reftable: add error related functionality
The reftable/ directory is structured as a library, so it cannot
crash on misuse. Instead, it returns an error code.

In addition to signaling errors, the error code can be used to signal
conditions from lower levels of the library to be handled by higher
levels of the library. For example, in a transaction we might
legitimately write an empty reftable file, but in that case, we want to
shortcut the transaction.

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00
Han-Wen Nienhuys
27f7ed2a07 reftable: add LICENSE
The objective of this code is to be usable as a C library, so it can be reused
in libgit2.

This is currently using a BSD license as it is the liberal license I could find,
but this could be changed to whatever fits the stated goal above.

This code is currently imported from github.com/hanwen/reftable. Once this code
lands in git.git, the C code will be removed from github.com/hanwen/reftable,
and the git.git code will be the source of truth.

Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-10-08 10:45:48 -07:00