Skip to content
This repository was archived by the owner on Jun 12, 2023. It is now read-only.

Conversation

@stavrospapadopoulos
Copy link
Contributor

New features

  • New read/write modes. Now a user can select to read/write cells in row- or column-major order within a specified subarray, rather than in the native array global cell order. See the tutorial and added examples on the TileDB official website.
  • Reading dense arrays with empty cells. Now the user can read a subarray of a dense array that overlaps with empty regions, and get back "empty" cells in addition to the normal non-empty values. See the tutorial and added examples on the TileDB official website.
  • Added a separate C API for syncing arrays and separate array attributes upon writing. See the C API tutorial on the TileDB official website.

Other changes

  • Added more unit tests.
  • Made MPI optional. See the Github Wiki to see how to build TileDB.
  • Changed the way verbose error messages are printed. See the Github Wiki for more details.
  • Now the fragment name carries also the MAC address of the machine that created it. This was necessary for deployment on a parallel file system, as different machines may produce the same fragment name based only on thread id and timestamp.
  • Various bug fixes.

stavrospapadopoulos and others added 30 commits October 24, 2016 17:19
…ted in native order or in column- or row-major order within the specified subarray
…for manual syncing an array or a specific attribute file.
…g on properly including -pthread in case OpenMP is disabled. Added also a new preprocessor flag USE_PARALLEL_SORT to explicitly state whether parallel sort will be used or not when OpenMP is enabled (up until now, parallel sort was always used when OpenMP was enabled).
stavrospapadopoulos and others added 29 commits November 9, 2016 18:52
…ent_cell_ranges in the case of multiple dense fragments and a subarray that falls in 'empty' space.
…ate::get_next_overlapping_tile_dense, (ii) fixed sort algorithm in multi-fragment reads, (iii) fixed calculation of number of cells in a cell slab in ArraySortedWriteState.
… address in addition to the thread id and timestamp (very important for the case TileDB is used on a parallel file system).
…te_state.cc, respectively (it has to be created before the mutex initializations).
…ated thread. Otherwise, the terminated thread does not release its resources. This can cause all sorts of problems (a typical error is that we will reach the OS limit for thread creation and we will not be able to create a new thread).
Added unit tests, added MAC address to the fragment name and bug fixes.
…the number of buffers due to the extra coordinates buffer needed for the sorting.
@stavrospapadopoulos stavrospapadopoulos merged commit 82ef689 into master Jan 3, 2017
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants