model-writtenrun 7342 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
May 2020 - Apr 2021 · 22 entries
- Daniel Gustafsson proposed NSS/NSPR as an independent TLS implementation with certificate databases and FIPS support, supplying frontend/backend code and backend-specific SSL test modules while identifying substantial unfinished work in passphrases, portability, documentation and auxiliary crypto consumers #1.
- Thomas Munro's early testing exposed OpenSSL-test and build problems #4; Daniel repaired the test harness and removed the already-landed statistics prerequisite #8, then extended sslinfo through the existing TLS abstraction #10.
- Andrew Dunstan supplied Windows/MSVC integration and portability fixes #11, followed by pre-generated certificate databases and successful OpenSSL and NSS buildfarm runs #12.
- Daniel introduced NSS documentation and a separate OpenSSL markup cleanup #16; Michael Paquier committed the cleanup as 089da3c477, adding line wrapping and one missing documentation occurrence #19.
- Daniel removed binary certificate databases that CI could not apply and implemented server passphrase callbacks inside forked processes, without OpenSSL's terminal-prompt fallback #23.
- The proposal gained pgcrypto and strong-random support, while SHA changes awaited a separate API redesign; Blowfish retained the internal implementation and CAST5 was omitted because NSS crashed on tested inputs #24.
- Andres Freund challenged socket errors, password cleanup, certificate-name validation, global state and the submission's size #25; Daniel fixed local defects #26 and split the work into six reviewable patches #29.
- Andres proposed a single TLS-library selector and discussed runtime selection for interoperability testing #28; Daniel adopted build-time --with-ssl={openssl,nss}, retaining --with-openssl as an alias #30.
- Andrew requested NSS-native certificate generation #31 and proposed cross-testing saved binaries from separate OpenSSL and NSS buildfarm animals #33, which Daniel supported #35.
- Heikki Linnakangas identified incorrectly shared password state #27, prompting connection-specific tracking and initial SCRAM channel-binding support #34; Jacob Champion's macOS testing then exposed unsupported shared-cache locking #36, leading to a non-shared session cache #37.
- Jacob supplied fixes for NSPR private-data assertions #40, nonblocking read timeouts and disconnect errno handling #42, and an infinite certificate-digest loop #44; Daniel integrated these into the full series and improved diagnostics and SCRAM testing #43 #45.
- After Jacob reported handshake failures leaking FATAL messages in cleartext #46, Daniel resumed work after the year-end gap with corrected error handling, NSS-native certificate generation and initial cryptohash support #49 #50.
- Certificate testing exposed missing usage extensions, which Daniel made explicit in the generation rules #52; Jacob then implemented common-name extraction that handled attribute order, escaping, missing names and embedded nulls #53.
- Michael urged extraction of independently useful infrastructure and reported cryptohash allocation, initialization and error-return problems #54, then posted a standalone configure-selector patch #57 that Jacob successfully built using both switch spellings #58.
- Jacob suggested --with-crypto #58, but Daniel preferred --with-ssl because PostgreSQL coupled the crypto choice to TLS support #60, and Michael accepted that name #62.
- Daniel separated the configure prerequisite and explored a static cryptohash context to reduce initialization costs #63; Michael combined that split with his earlier work, corrected pgcrypto selection and remaining test references, completed the documentation and committed fe61df7f82 #67.
- Daniel rebased the remaining NSS series, restored Jacob's accidentally omitted common-name changes and revised initialization and shutdown handling #68; Jacob added a missing root CA and enabled the client-certificate and SCRAM tests #74.
- Daniel further split the series and moved shared implementation out of headers #78; Michael reviewed its independent SSL/OpenSSL cleanup #80 and committed it with light comment edits as 092b785fad.
- The remaining series improved SCRAM and documentation #86, while full sslinfo distinguished-name parity was deferred pending better tests #85; subsequent fixes added CRL-directory compatibility and named test-harness parameters #89, repaired OpenSSL regressions #91 and completed missing MSVC definitions #92.
- Stephen Frost's detailed review found broken cipher selection and questioned unnecessary callbacks, types, naming and documentation #93; Daniel addressed these across several revisions, including removal of redundant I/O callbacks and the frontend handshake loop #94 #105 #113.
- Jacob's investigations exposed initialization and multiple-database problems #96 #104, then showed that implicit NSPR initialization was not thread-safe and relied on a special primordial thread #110; Michael and Stephen confirmed that libpq could not assume a main-thread caller #111 #112.
- Michael supplied another rebase and found failing distinguished-name mapping #115, then judged the work premature for PostgreSQL 14 #116; after parental leave, Daniel returned with DN resolution, frontend locks, explicit database token names and certificate-selection tests #117 #118.
May 2021 - Feb 2022 · 18 entries
- A parallel terminology discussion favored retaining established SSL names in code for compatibility and familiarity, while improving explanatory SSL/TLS documentation #119 #124 #129.
- Jacob repaired a missing CRL-directory test database #130, which Daniel integrated #131, then traced NSS shutdown failures to leaked references and introduced shutdown diagnostics, resource cleanup and an empty-stderr assertion in connect_ok #132.
- Daniel identified the need to clear the NSS session cache #133; Jacob revised the fixups #134, and Daniel incorporated them with close_notify tracking #136.
- Further testing found cache-clearing crashes, NULL-host crashes and misleading certificate and EOF diagnostics #137; Daniel folded those fixes into the series with tentative SNI handling #138.
- Jacob requested a way to discover libpq's TLS implementation without opening a connection #140; Michael favored broader security-library discovery #141, while Daniel preferred extending existing attribute queries and separating the broader capability discussion #142.
- After maintenance rebases, Jacob corrected SNI/IP behavior and honored sslsni=0 #148; Rachel Heaton's testing exposed an enum mismatch #149, which Daniel fixed #150.
- Jacob traced forbidden exit references to frontend-unsafe allocation helpers #152 and replaced them #154, accidentally starting from the version before Daniel's enum correction; Daniel merged both branches in the next full series #155.
- Daniel refreshed the series for Perl and SSL-test changes and added NSS/NSPR version checks #159; Joshua Brindle's Fedora testing drove include-path fixes #161 #162 #163, while pgcrypto adaptation explicitly left PGP ElGamal and RSA unsupported because NSS lacked equivalent bignum functions #164.
- Joshua reported a crash when connecting before creating the client certificate database #170; Daniel corrected stale connection state #171, and Joshua confirmed the crash disappeared #172.
- Joshua subsequently reported teardown errors and an invalid-database hang #175 #176, while Jacob identified unreachable certificate-name code #177; Daniel repaired cleanup and locking and simplified the code #179.
- A generated-configure omission still broke LLVM builds #181; Daniel corrected it, added cryptohash error reporting and experimental NSS CI #183, then repaired OpenSSL key-handling regressions in the refactored tests #185.
- Jacob reported NSS slot exhaustion after repeated context creation #187 and supplied a reproducer failing after roughly 98 connections #191; the supplied thread shows no subsequent resolution.
- Andres's broad review questioned post-fork initialization, delayed validation, thread safety, duplicated code and patch size #189; Daniel applied local fixes, removed a questionable shutdown callback and proposed independent test changes plus a possible client/server split #192.
- Andres clarified that the split should permit incremental commits and requested connection-speed testing #193, leaving initialization validation, some locking questions and the final integration strategy unfinished.
- Concern about NSS documentation and other projects abandoning support led Daniel to propose rejecting the main series #198; Andres softened his objection #200, while Stephen defended NSS's maintenance and licensing and disputed that the work should be discarded #199 #204.
- Peter Eisentraut questioned NSS's remaining benefits after OpenSSL 3 #207 #211; Daniel defended alternative libraries and certificate stores #208, and Robert Haas said adoption could be reasonable with committed maintenance and buildfarm coverage #212.
- The supplied discussion ends without an NSS landing or an uncontested rejection: restored documentation improved the outlook, but Daniel still identified missing documentation for important APIs such as NSS_InitContext #217 #218 #221.
- After the final supplied message, Jacob's independently useful connect_ok stderr assertion landed as 4d373e0528 through the separate discussion thread #45445, whose matching submission #498241 superseded this thread's earlier inclusion #192; obsolete test-count adjustments were omitted, and the main NSS implementation remained uncommitted in the supplied record.