Mar 2019

7 commits
Allow fractional input values for integer GUCs, and improve rounding logic.
11 Mar 2019Tom Lanemaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 100% suremodified before commitlight changesfrom patchset #40

Combines both attachments from #40; the guc.c changes and reloptions regression changes match exactly. The committed config.sgml adds an explanation of rounding to the next smaller unit and slightly edits the integer-type wording.

Reduce the default value of autovacuum_vacuum_cost_delay to 2ms.
10 Mar 2019Tom Lanemaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #33

Extracts the 20ms-to-2ms default change from the second attachment in #33, updating guc.c, postgresql.conf.sample, and config.sgml. The floating-point infrastructure was committed separately.

Revert "Increase the default vacuum_cost_limit from 200 to 2000"
10 Mar 2019Tom Lanemaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 morereverts 1 commit here
LLM analysis, 100% surewritten from the discussion, no matching patch

Implements the revert proposed in #25 and explicitly required but omitted from the attachments in #33. It restores vacuum_cost_limit to 200 in the same four files changed by the original patch.

Convert [autovacuum_]vacuum_cost_delay into floating-point GUCs.
10 Mar 2019Tom Lanemaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 99% surepart of a posted patchsetmoderate changesfrom patchset #33

Combines the floating-point unit infrastructure and vacuum-delay conversion from #33, leaving the default reduction for a separate commit. It adds the catalog-version bump, documents microseconds, and changes convert_real_from_base_unit() to tolerate relative rounding error within 1e-8.

Include GUC's unit, if it has one, in out-of-range error messages.
10 Mar 2019Tom Lanemaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 100% surewritten from the discussion, no matching patch

Implements Tom Lane's unattached proposal in #35 to include converted units in out-of-range errors. It adds get_config_unit_name(), uses it for integer and floating-point errors and pg_settings, and updates the regression expectation.

Disallow NaN as a value for floating-point GUCs.
10 Mar 2019Tom Lanemaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+6 more
LLM analysis, 100% surewritten from the discussion, no matching patch

Addresses the NaN acceptance problem identified in #33, which explicitly says the attached patches do not fix it. The commit adds an isnan() rejection to parse_real() and regression cases for NaN, infinity, and an out-of-range delay.

Increase the default vacuum_cost_limit from 200 to 2000
6 Mar 2019Andrew Dunstanmaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 100% surelanded as postedfrom patchset #51 of 1 reviewer saw the final version

All four default-value substitutions match #5 exactly, with identical added and removed lines. Andrew Dunstan confirms the landing in #17, and the commit credits patch sender David Rowley as author.

  • Andrew Dunstansaw the final versioncommittercommitted the patch

Credits across 7 commits

Author
Tom LaneLLM3
David Rowley1
Reviewer
Andrew Dunstancommitter1
Committer
Tom Lane6
Andrew Dunstan1

N LLM means N of the credits come from LLM analysis, not from commit trailers.