Skip to content
This repository has been archived by the owner on Jul 19, 2021. It is now read-only.

Build fixes #2

Merged
merged 2 commits into from
May 9, 2017
Merged

Build fixes #2

merged 2 commits into from
May 9, 2017

Conversation

joakim-hove
Copy link
Contributor

No description provided.

@joakim-hove joakim-hove closed this May 9, 2017
@joakim-hove joakim-hove reopened this May 9, 2017
@joakim-hove joakim-hove merged commit 29b9059 into equinor:master May 9, 2017
joakim-hove added a commit to joakim-hove/libres that referenced this pull request Nov 9, 2017
# This is the 1st commit message:

Reduced importance of enkf_state:

The main purpose of this commit is to enable a dynamic number of
realisations when simulating, to achieve this it has been essential to
reduce the importance of the enkf_state implementation. The enkf_state
object was originally introduced as an object to represent one
realisation, and the enkf_state object would have application
lifetime. We have refactored quite a lot, generally shifting in this
importance from the enkf_state object to:

 1. enkf_fs - i.e. the storage system. We do not really care/trust the
    values held in memory of the enkf_state object, instead we
    immediately query the file system when we need data.

 2. Configuration, configuration and configuration. Meaning that the
    configuration information of various kinds is much more important,
    using this configuration information realisations can be created on
    demand.

In addition the ert_run_context and run_arg datastructures have gained
increased importance. The important point about these datastructures is
that they have lifetime corresponding exactly one simulation batch.
Observe that in the refactoring process some old-untested functionality,
typically from the EnKF era, has been completely removed.

  1. The member_config object has been completely removed; there is
     temporary a __iens member in the enkf_state object. But most use of
     iens should come from run_arg now.

  2. The rng member has been removed from the enkf_state object, instead
     an rng is passed explicitly when required.

  3. The subst_list has been removed from the enkf_state object. Instead
     a subst list is instantiated when needed in the export
     function. Observe that as part of this process the magic strings
     related restart files and timesteps have been removed.

  4. The jobname and eclbase settings have been merged, internally only
     the jobname is used.

  5. The run_arg datatype has got a new element job_name.

  6. The callbacks from the queue system are implemented with a new
     callback_arg type. This type can invoke the callback functionality
     without explicit reference to a enkf_state object.

  7. Have removed the CASE_TABLE functionality completely.

  8. Have changed the config settings ENSPATH, RUNPATH and DATA_ROOT to
     be extracted as config_iget_as_path( )

  9. Have as part run_arg implementation forced load_start to step1,
     even for step1 == 0. This might have issues for summary files with
     missing data for step 0?? This is to match the default convention
     of finding GEN_DATA results at step == 0.

  10.The ert run context will now have run_arg 'slots' for the full
     size, with NULL for inactive elements. The result should be 'flat'
     indexing without ensemble awareness.

# The commit message equinor#2 will be skipped:

#	fixup: Added test for run_context is_active method.
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.

1 participant