NEWS

teems 1.0.1

macOS (Docker Desktop): a second model run into the same folder within a second of the previous one fails with "cannot open sol.var"; stale solution files are now removed inside the solver container

ems_solve(): verbosity sets the console detail only; the full solver log always goes to model_diagnostics.txt. The default console shows the phase lines and the accuracy figures table; per-step timings, memory probes, convergence codes, the per-subinterval and updated-data accuracy summaries and the Zerodivide warnings stay in the log, and Zerodivide substitutions are reported once with a count
The run summary shows - for a convergence code without a lowest-figures value (judged-zero results), not an empty cell
Example scripts: the full uniform population shock object is pop_shk (it was misnamed numeraire)

ems_solve()/solve_in_situ(): solution_method defaults to "Gragg" (was "Johansen"); "mod_midpoint" is renamed "Gragg" (it was always effectively Gragg); steps defaults by method
laA, laD, laDi and append_args are removed as named arguments; the MA48 workspace sizes and expert solver flags are validated ... arguments
ems_model(var_omit = ) is removed; in-TAB Omit statements are ignored with a message
GTAP-INT is removed; GTAP-RE is replaced by version 2 on the GTAPv7.1 core
Element names are case-insensitive on input and lowercase in all outputs
teems checks the solver image version before solving and aborts on a missing version, a major-version mismatch or an image below 1.1.0

ems_probe(): structural probe of a deployed model; recommends a matrix method and resources and gives a condensation verdict
ems_RK(): Runge-Kutta solution methods (RK2, Heun, RK4, BoSha32, DoPri54) with adaptive step control

Vetted models: GTAPv7.1 (updated from GTAPv7.0), GTAP-AEZ, GTAP-E, GTAP-EP (GTAP-Power)
ems_data() loads a single-HAR database (par_input/set_input optional)
ems_data(par_weights = ): "value" (default, the FlexAgg rule) or "share" weights for parameter aggregation, overridable per parameter
Closure swaps were applied ins-before-outs, rejecting valid sequences; two swaps in on one tuple wrote duplicate closure entries
ems_scenario_shock(partial = TRUE): a scenario may cover only some of the aggregated tuples; the uncovered ones keep a zero change, and an aggregated tuple covered in part stops the deployment by name

solution_method gains "Euler" and the Runge-Kutta family; Euler and Gragg accept a single step count (no extrapolation), and Gragg accepts all-odd counts
Condensation: in-TAB Substitute/Backsolve statements and ems_model(backsolve = , ignore_condense = )
ems_solve() gains n_threads, precision = "double" and verbosity
Run-mode switches (assertions, range tests) are set with ems_option_set()
Coefficients are read from the solver's binary dump at full precision (previously CSVs with six fixed decimals); ems_deploy(write_coefficients = TRUE) restores the CSVs
Solver failures are reported by class (resource, size, model specification, interface) instead of a generic exit status

Set expressions, conditional set builders, $POS, Mapping, levels variables and equations, Complementarity, PostSim, IF terms in formulas and equations, Read (IfHeaderExists)
Malformed TABs abort by name at model load instead of inside the solver

ems_uniform_shock() on a one-index variable restricted to an element or subset shocked the wrong elements (element 0 / the first n elements of the full set) (solver 1.1.0)
DPSM (all GTAP models) was summed over aggregated regions, so an aggregated region held the number of subsumed original GTAP regions instead of 1
GTAPv7: the scalar pxwwld was dropped from E_c1_cr, E_cnttotr and E_cntpinv because the solver did not recognize a variable followed by a closing bracket (solver 1.1.0)
DBBD solves left residuals that multi-step runs accumulated; every DBBD solve now takes one refinement step (solver 1.1.0)
Data rounding used fixed decimals, writing small values (below 5e-7) as 0; values now keep ndigits significant digits
Whole-valued Real data at or above 2^31 was written as empty fields
Sets read from a header, and explicit set lists, were written in alphabetical rather than file/declaration order, misaligning data indexed over unions and changing $POS arithmetic (no effect on the shipped GTAP models)
Custom and scenario shocks with duplicated or overlapping rows were accepted and could shock elements never specified
Redeploying within the same minute appended to the existing shock file
A user shock file with an argument-less uniform shock shocked only the variable's first element (solver 1.1.0)
A uniform shock on a scalar variable declared without sets was written as Shock phi(NA)
GTAP-RE: the PVALFWDTIME discount did not scale with time-step length

On Linux the solver container runs as the calling user (previously failed for users other than uid 1000)
Deploy paths containing spaces work
A deploy directory accepts one solve at a time
Ctrl-C stops a running solve: the solver container is removed and the run is reported as interrupted (previously the solve ran on and was reported as a failure)

write_dir removed from ems_deploy(), default write directory now tempdir() and can be overridden via tempdir arg in ems_option_set()
Explicit path first arg for ems_example(), with no default
ems_example() now accepts outputs from GTAP_convert()
GTAP_convert() inputs now har-specific
Expanded and more descriptive examples