Jun 2021

1 commit
Rename PQtraceSetFlags() to PQsetTraceFlags().
11 Jun 2021Noah Mischmaster14beta2REL_19_STABLE14beta2REL_18_STABLE14beta2REL_17_STABLE14beta2REL_16_STABLE14beta2REL_15_STABLE14beta2REL_14_STABLE14beta2
LLM analysis, 100% surewritten from the discussion, no matching patch2 of 2 reviewers saw the final version

Noah proposes PQsetTraceFlags in #230 and announces the rename in #233; no attached patch implements it. The commit updates the function, export, public declaration, documentation and test caller.

  • Alvaro Herrerasaw the final versionLLM analysis, 100% sureThe prose in d0e750c0ac explicitly credits him as a reviewer, alongside his naming discussion in #231.
  • Robert Haassaw the final versionLLM analysis, 100% sureThe prose in d0e750c0ac explicitly credits him as a reviewer, independently of his brief support in #232.

Mar - Apr 2021

9 commits
Suppress length of Notice/Error msgs in PQtrace regress mode
9 Apr 2021Alvaro Herreramaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surelanded as postedfrom patchset #2271 of 1 reviewer saw the final version

Matches #227 modulo rebase: fe-trace.c prints NN for backend Notice/Error lengths in regression mode and updates the same two expected trace files.

  • Tom Lanesaw the final versiontrailerHe reviewed the first patch and its explanation in #222 and #225, then accepted the revised comment in #228.
Strip file names reported in error messages on Windows, too.
2 Apr 2021Tom Lanemaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surewritten from the discussion, no matching patch

Tom supplied the inline elog.c backslash-stripping fix in #222 and announced its commit in #223. The committed form strips backslashes after slash processing and adds an explanatory comment.

Fix setvbuf()-induced crash in libpq_pipeline
1 Apr 2021Alvaro Herreramaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surewritten from the discussion, no matching patch

Implements Tom's Windows buffering diagnosis in #211 and shift-expression findings in #208 and #210, using PG_IOLBF and 1LL << 62 in libpq_pipeline.c; no matching attachment exists.

libpq_pipeline: Must strdup(optarg) to avoid crash
1 Apr 2021Alvaro Herreramaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surewritten from the discussion, no matching patch

Follows Alvaro's argument-lifetime theory in #206, adding strdup calls and a -r option in libpq_pipeline.c without an attached patch. Tom's skeptical response #207 came after the commit.

Remove setvbuf() call from PQtrace()
31 Mar 2021Alvaro Herreramaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surewritten from the discussion, no matching patch

Implements Tom's objections in #204 by moving setvbuf from PQtrace into libpq_pipeline and opening the trace with mode w; Alvaro confirms these changes in #205.

Initialize conn->Pfdebug to NULL when creating a connection
31 Mar 2021Alvaro Herreramaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surelanded as postedfrom patchset #2011 of 1 reviewer saw the final version

Exactly matches the one-line conn->Pfdebug = NULL addition in #201. Tom notes in #203 that existing MemSet already initializes it, so this improves explicitness rather than fixing the suspected crash.

  • Tom Lanesaw the final versionLLM analysis, 100% sureIn #203, eleven seconds before commit, he inspected #201 and identified the existing MemSet that made its proposed crash explanation incorrect.
Disable force_parallel_mode in libpq_pipeline
31 Mar 2021Alvaro Herreramaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surewritten from the discussion, no matching patch

Tom diagnosed force_parallel_mode timing differences in #197, and Alvaro described disabling it in #198. The commit also moves lc_messages setup outside tracing and adjusts expected output, without an attached patch.

libpq_pipeline: add PQtrace() support and tests
30 Mar 2021Alvaro Herreramaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 99% surepart of a posted patchsetmoderate changesfrom patchset #1830 of 1 reviewer saw the final version

Takes the pipeline tracing and comparison-test components of #183, leaving the core feature to its separate commit. It adds regression-mode flags, three trace cases and revised TAP file-reading/error handling.

  • Iwata, Ayasaw an earlier versionLLM analysis, 98% sureShe explicitly reports reviewing and successfully regression-testing #183 in #191, before the additional trace cases and harness changes.
Improve PQtrace() output format
30 Mar 2021Alvaro Herreramaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #1911 of 10 reviewers saw the final version

Derives from #191, with libpq-trace.c renamed to fe-trace.c, separate timestamp/regression flags, primitive-level NNNN/SSSS redaction and shared Notice/Error handling. Alvaro describes these committer edits in #192.

  • 綱川 貴之 (Takayuki Tsunakawa)saw an earlier versiontrailerHis extensive ABI, allocation, cursor and protocol reviews culminated in #178 and #180, before substantive cursor and regression-mode changes.
  • Kirk Jamisonsaw an earlier versiontrailerHe tested builds, COPY, timestamp suppression and pipeline traces, reporting remaining duplicate-message defects while supplying fixes in #139, before the architectural rewrite.
  • Kyotaro Horiguchisaw an earlier versiontrailerHe reviewed interfaces and cursor logic, reproduced an SSL-reconnection crash in #127, and reviewed the rewritten decoder through #175 before later substantive changes.
  • Tom Lanesaw an earlier versiontrailerHe challenged the duplicate logger in #39, required whole-message decoding in #134 and reviewed trace placement in #146, before subsequent decoder and regression-mode revisions.
  • 黒田 隼人 (Hayato Kuroda)saw an earlier versiontrailerHe identified the unavailable Windows setlinebuf call and proposed a portable replacement in #76, before the later tracing rewrite.
  • Nagaura, Ryoheisaw an earlier versiontrailerHe tested the original logger and identified timestamp, size-reporting and unwanted-output issues in #25, before both major redesigns.
  • Ryo Matsumurasaw an earlier versiontrailerHis detailed review in #73 exposed crashes, lookup-table errors, field-ownership mistakes and binary-output defects before substantial reworking.
  • Greg Nancarrowsaw an earlier versiontrailerHe tested tracing through psql and reported allocation-failure, binary COPY and terminator problems in #68, before the architectural rewrite.
  • Jim Dotysaw an earlier versiontrailerHe built and tested the first logger, including directory failures and performance measurements in #13, before its replacement by the PQtrace redesign.
  • Álvaro Herrerasaw the final versiontrailerHe repeatedly reviewed, tested and revised the implementation, described final edits in #192, and reviewed the committed version as its committer.

Mar 2021

1 commit
Re-simplify management of inStart in pqParseInput3's subroutines.
11 Mar 2021Tom Lanemaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1REL_13_STABLE13.3+4 more
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #149

Starts from #149, centralizing inStart advancement and fixing malformed ParameterDescription handling in fe-protocol3.c. It additionally removes redundant consumption checks from three parsing helpers, as Tom explains in #151.

Credits across 11 commits

Author
Alvaro Herrera2 LLM3
Iwata, Aya1
Tom LaneLLM1
Reviewer
Tom Lane2
Alvaro Herrera1
Greg Nancarrow1
Hayato Kuroda (Fujitsu)1
Jim Doty1
k.jamison@fujitsu.com1
Kyotaro Horiguchi1
Matsumura, Ryo1
Nagaura, Ryohei1
Tsunakawa, Takayuki1
Committer
Alvaro Herrera8
Tom Lane2
Noah Misch1

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