DIALS core meeting 2021-07-15

[Previous meeting: 2021-07-01]

Previous Actions

• MG: Giving a typing intro lecture on 14th of July, 13:00 BST
• dx2 proposal
  • MG to put in a PR to create a proposal in the proposal-space
  • MG+ND: Come up with a proposal to move away from all code being in header files and consolidate into a single library
  • Action: Agreed the process was useful. Shelve and come back in three months, see what everyone’s time looks like
• Merge dxtbx src/ layout change (prev)
• ND: conda-forge pycbf: Make new release to use dials-data directly for tests

Agenda

Status checks before merging on dials/dials

• Agreed previously on 2021-06-17
• Review planned for 2021-07-15
• What have peoples experiences of this been?
  • Everyone happy

“Contiguous Nexus” PR

• cctbx/dxtbx#356
• What is status of this?
• Raised internally whether this would benefit from flumpy conversions? Is numpy bypassed entirely already?
• Action: Non-draft pending checks against the possible issues that @dwpaley has raised, and checking for no-contiguous cases

cbflib conda-forge package/pycbf

• Outstanding: dxtbx#368 for optional usage in dxtbx, no binary dependence on cbflib/_adaptbx if present.
  • Still stray issue with dataset SPring8_ADSC_SN916 being the only one requiring FormatCBFFull. One possible solution is adding a special Format: dials/dials#366
• Still need to test with labelit/labelit_regression once the test suites actually work
  • All except one labelit tests are now working. On NKS todo list to fix.
• CBFlib 0.9.7 work is ongoing at cbflib/cbflib#1. Tests now pass.

src/ layout for dxtbx

• Once 2021.06 July release is out cctbx/dxtbx#382 Merged.
• Outstanding issue: cctbx/dxtbx#393 - dispatcher re-export of non-dxtbx tooling appears not to work on Windows
• Action: Ask CCTBXBB if anyone has knowledge or experience of this behaviour

flumpy – flex/numpy bridge

• cctbx/dxtbx#377 Merged
• Tests failed on Windows: cctbx/dxtbx#392 - Flex doesn’t bind C long long on any platform, but On POSIX platforms long and long long are both 64 bit integers, and numpy tends to pass "q" codes for any 64-bit integer dtype. On Windows, flex still doesn’t bind long long but does bind int64_t (64-bit unsigned is bound as size_t) which matches the "q" dtype.
• Additional problem: cctbx/dxtbx#398 from_numpy currently fails if scitbx.array_family.flex has not already been imported.

DIALS versions in Phenix releases

• Short discussion about the development branch of CCTBX pointing to old dials builds, is all clear now

Nonconventional P1 indexing

• D: Observed that indexing in P1 with b>c in a non-conventional settings appears to break (in stills_process, dials.index to be checked)
• stills_process might not process triclinic cells in an unconventional settings - possibly tries to order base vectors?
• Should this be indexing in the lattice that you give it, rather than changing the setting?
• Action: Dan to try and demonstrate this on a simple/lysozyme example and file an issue

Next meeting

Thursday, July 29th, 4pm UK (BST), 8am PDT.