
The database build a run reads, and the check that it is the evidence's
Source:R/build_record.R
knw_build_record.RdA grade or a comparison reads four things from the database besides the
FWA tables: link's persist schema, the run that built it
(<schema>.log_recompute's run_uid), wet's flow load (the table comment,
knw_discharge_load()) and the walk installed from this package
(knw_walk_md5()); fresh's version decides the comparison's regeneration.
0320 and 0330 read this record at connect and refuse a schema whose run
is not the evidence ledger's, before any work, so the same command runs on
any machine whose database holds the build and stops on one that does not
(#87). A provincial run in link renames fresh_default and builds a new one
in its place (link#361), which is the case the check catches.
Value
A list: schema, run_uid (the schema's runs, ;-joined),
discharge_load, walk_md5, fresh_version, tokens (link's output
vocabulary, c(cut =, acc =, grad =, flow =), read from the build's rows
of <schema>.log, where link writes them; each NA where that log predates
link 0.70.0 and carries none), log_run_uid (the runs whose log rows
were read, ;-joined), log (which rows those are, as a clause a reader
of log appends) and line, the record as one line for a log. A
schema whose log holds two values on one axis, or rows with a token beside
rows without one, is refused: it is not one build. The tokens are read over
the log rows of the runs log_recompute names, because a provincial run
leaves the previous run's log rows in place (#145), and over every log
row where none of those runs is in log, because a build's log and its
recompute's log_recompute can carry different run ids (link's four-group
proofs), where it must hold one run. A run set aside that was logged after
the build began is refused. The line names any run whose rows were set
aside, and, where every row was read, the log run the tokens came from.