History / Migrating the Database Backend

Ten more passages of this wiki's own text, on seven pages, say what changes where the service may not write hMailServer.ini, as ServiceAccount.md's table of the writes to hMailServer.INI says with batch51 on master, 5e49abf06, and IniSettingStore has done since the tag v6.3.4, where they said without that condition what the server writes into the file, or that an edit to it is undone, the shape of the passages the previous commit corrects. Migrating the Database Backend's own sentence on the stored settings under Rolling back says that the rolled-back server rewrites the file's [Settings] section to match where the service may write the file. Warm Standby's worked commands say that a value typed into the INI is not used once the key has a stored value, where they said it was undone. Settings Overview says that DELETE /api/v1/settings/ini/{name}, hmctl ini delete and Settings.DeleteIniSetting remove the line and the row together where the service may write the file, and that where it may not and the file has the line the line stays and the row is kept holding the key's default, which the server treats the same, or the removal is refused for PasswordPepper and a key this build does not read, and that the server adds the [SettingsStore] section where it may write the file. Settings Reference's section 4a says the same of a removal, a dropped row being put back from the line at the next start and the removal said in the application log (IniSettingStore::RemoveSetting), and that what a file the service may not write changes, which the tag v6.3.4 already has, is read from the tree at 5e49abf06, where the section named 3233b0cc4 alone, a tree from before it. Installing on Linux says that on a package installation, whose file is 0640 root:hmailserver, a value stored over the API is in effect and the line left as it was is not an error, the application log saying, where application logging is on, that the file is read-only to the account the server runs as and that its [Settings] copy is not refreshed (IniFileSettings::InitializationFileWritable, IniSettingStore::WriteSetting and Synchronize), where it said that such a value was reported as HM5806 when it was stored and as HM5803 at every start; the packaging README at 5e49abf06 still says so under What lands where, and this wiki follows the code and ServiceAccount.md, which says of the Linux packages that a setting saved is no longer an error at every start. Changes Since 6.2.10's 6.3.4 entry on the settings store and its note on the closed table say that the server writes the [SettingsStore] section where it may write the file, and Upgrading hMailServer's note on rolling back from 6.3.4 says that the newer server writes the [Settings] secrets into the file as HMK1: envelopes where it may write the file. Signed-off-by: Christopher Holloway <chris.holloway@progressiverobot.com>

Christopher Holloway committed Oct 21, 2026

Four passages of this wiki's own text say what the code and the documents say, where they said otherwise before batch50 and still did after it: Relocating an Installation changes the stored [Settings] paths in step 9, once the moved server is running, or through a [SettingsOverride] section at its first start, where its own route stored them "before step 2, while the old server is still running" and said they took effect at the next start - each stored value is applied at once, but for the listeners' certificate paths and the ACME certificate directory, which wait for a reinitialise, so that route pointed the running old server at a tree not yet copied; the same passage says the two trust files are not among the rows that would come back pointing at the old tree, and that from 6.3.6 the live update's trust is read from the file alone like PasswordPepper, with UpdateTrustRootsFile and UpdateLogPublicKeyFile changed there in step 4, where it called PasswordPepper the one exception; the Regression Test Environment's table names pre-flight check 8, the protocol listeners, for a missing SMTP listener, where it named check 11, the orphan test files (build/preflight-tests.ps1); and Migrating the Database Backend and the Upgrading Guide say that a database password re-entered on another machine is written protected by DBSetup.exe, or as a plain value with PasswordEncryption=0, which the server reads as it is and does not protect again, where they said the server would re-protect a plain value (IniFileSettings reads the password by its PasswordEncryption type, and only the COM Database methods DBSetup uses write a protected one). Signed-off-by: Christopher Holloway <chris.holloway@progressiverobot.com>

Christopher Holloway committed Oct 20, 2026

Migrating the Database Backend's comparison table says what a failed schema upgrade leaves on SQL Server Compact as Upgrading.md says it with batch50 on master, 0c0b6bae7, and as this wiki's Upgrading Guide has said since 9a56d55: nothing applied, but for a chain from a database older than 6072 that crosses 6107, whose steps before 6107 DBUpdater commits so that 6107's greylisting keying runs outside the transaction, and which a failure from 6107 on leaves at 6106, where the cell, written in 90a1f71, said nothing applied with no exception. The page's header says that the table makes that one exception, which the document's PostgreSQL row leaves out, where it said that the table says the same as the document. Signed-off-by: Christopher Holloway <chris.holloway@progressiverobot.com>

Christopher Holloway committed Oct 20, 2026

Migrating the Database Backend has the document's em dashes again, thirteen of them in the twelve lines it copies unchanged from MigratingDatabaseBackend.md as master has it from batch50, 0c0b6bae7 - the opening's, the three of Why this works, the line after the comparison table, the check that every message is inside the data folder, the two of step 1, the archive's one file in step 2, the message that opens under Verifying it worked, the rollback, the changed ids and the moved data folder - where it had written them as hyphens since the page was rewritten on 8 September 2026, the copy of 24 September 2026 having brought back the other four. Its header names this wiki's own additions, where it named only the comparison table under Choosing a target and the line under it that points to Database TLS: the flowchart of the procedure; the diagram of the ids; the note that a restore refuses an archive taken by a newer version, the table of the [Database] keys and the note on the machine-bound password, under Before you start; the sentence on the backup's log and its verification, the table of the combinations a restore refuses, which stands in for the document's one example of them, the note on BackupMessagesDBOnly on 6.3.3 and the paragraph on why the database-only mode is close to mandatory on a migration, under step 2; the two lines of step 4's script that show what the archive holds, the diagram of what the restore does and the note on the staged directory; the command that counts each domain's accounts, aliases and lists, and its output; the sentence on the stored settings under Rolling back; the notes on the restore emptying the data directory, on the password on another machine and on InnoDB under Known sharp edges; and the pointer to moving to a new machine. It also says that the paragraph on where a message file lives is this wiki's wording of the document's, giving the public-folder directory as the code names it, #Public - IMAPConfiguration::GetPublicFolderDiskName, which PersistentMessage::GetFileName combines with the data directory - where the document at 0c0b6bae7 says .Public_Folder. So the copy is now the document's text but for its links, the arrangement of the comparison table and what the header declares, as 90a1f71 said each copy was; the document's two other em dashes are in its table of targets, in the MySQL row, and in its one example of a refused restore, which the page gives in its own arrangement and its own table. Signed-off-by: Christopher Holloway <chris.holloway@progressiverobot.com>

Christopher Holloway committed Oct 20, 2026

The eleven copies whose documents batch50 changes - the Assurance Case, Governance, the Release Process, the Security Policy, Regulatory Scope, Command Line, Event Scripts, Relocating an Installation, the Regression Test Environment, The Service Account and Migrating the Database Backend - are copied again from the documents as master has them from batch50, 0c0b6bae7, and each header says so and what the document changes, where they were the documents at batch49, f14412d39, or at v6.3.7, which batch49 left as they were. Migrating the Database Backend: step 3 runs DBSetup.exe - the wizard's Create a new hMailServer database, or DBSetup.exe /unattended with the HM_DB_* variables and HM_ADMIN_PASSWORD - which creates the database and its schema, writes the [Database] section through the service, the password encrypted (PasswordEncryption=6), and restarts the server on the new database, the login allowed to create a database and the database not existing yet, with Not DBSetupQuick.exe and the refusal of DBSetup under NT SERVICE\hMailServer, where it said to create an empty database, point hMailServer.ini at it by hand and run DBSetupQuick.exe, which made no usable database there; the short version, the MySQL and MariaDB section's account of who creates the database on Windows and on Linux, and step 4's first open of the new database follow it; the PostgreSQL row says that SQL Server's, PostgreSQL's and SQL Server Compact's DDL is transactional and that only MySQL and MariaDB commit each step's DDL as it runs, where it called PostgreSQL the only one of the four; and the note after the restore says that the schema is created in no transaction on every backend, where it said so of MySQL and SQL Server Compact alone. This wiki's comparison table under Choosing a target now has SQL Server and SQL Server Compact transactional and a failed upgrade leaving nothing applied on them, where it had SQL Server mostly transactional and SQL Server Compact committing as it goes, and its flowchart of the procedure has one step for DBSetup between the backup and the restore, where it stopped the service, created an empty database, edited the ini file, ran DBSetupQuick and started the service again; the table's link to DatabaseTls.md goes to this wiki's copy, Database TLS. The Service Account: in Reinstalling into a new folder, step 2 changes the settings file's other paths into the old installation - PostgreSQLSslRootCert and TlsCaFile, UpdateTrustRootsFile and UpdateLogPublicKeyFile, a path in [SettingsOverride] - to copies where nobody but the administrators may change them and deletes the lines the site did not set, and step 4 keeps the old folder private to SYSTEM and the Administrators until it is deleted; the live update's internal tests name 6.3.1 and 6.3.3 to 6.3.5, where they said 6.3.1 to 6.3.5; and its link to RelocatingAnInstallation.md under the message files outside the data folder goes to this wiki's copy, as its two others do. Relocating an Installation: step 1's inventory and step 4 name TlsCaFile beside PostgreSQLSslRootCert, and Seeing where the paths are says that Installation paths lists every path-shaped setting but that one. The Regression Test Environment: the list of what a new Windows bench needs names the release installers build/fetch-installer-fixtures.ps1 fetches for AdminUpdateTester and AdminUpdateStageTester. Command Line: the two queue examples filter on tries, the field a queue item has, where they named retry_count. Event Scripts: the rule action is "Run script function", as the Control Panel lists it, where it said "Run function". Regulatory Scope: the releases up to 6.3.3 whose files GitLab has are re-published there, 6.2.0 and 6.0.0-B4 being notes-only; SBOMs from 6.2.3, the five earlier releases that carry an installer having a CycloneDX SBOM of it alone; the assets from 6.2.19 to 6.3.3 signed but for 6.2.22-pre3's two SBOMs; and the releases before 6.3.1 unsigned because their original files are published byte for byte, where it gave a published release's immutability as the reason. The Assurance Case, Governance and the Release Process: GitHub's signer signed the releases from 6.2.19 to 6.3.3, where they said the releases up to 6.3.3. The Security Policy: Verifying a release names 6.2.22-pre3's two SBOMs and 6.3.4 beside the withdrawn and superseded pre-releases as the assets from 6.2.19 on without a Sigstore bundle. A link of the copied text to a document the wiki copies is a wiki link, and every other link to a repository file points at it as 0c0b6bae7 has it, where they pointed at f14412d39 or at v6.3.7; the text is otherwise the document's, but for the adaptations and additions the headers declare. The wiki's own tables on these pages say what the documents now say: the Assurance Case's traceability row for anyone downloading has the Sigstore signatures from 6.2.19 to 6.3.3 and the SBOMs from 6.2.3, and the Security Policy's row on the supply chain among the standing controls has 6.2.22-pre3's two SBOMs carrying no bundle and the SBOMs from 6.2.3, the five earlier releases that carry an installer having a CycloneDX SBOM of it alone, made afterwards. Signed-off-by: Christopher Holloway <chris.holloway@progressiverobot.com>

Christopher Holloway committed Oct 20, 2026

The twenty-two pages that copy a repository document are copied again from the documents as the tag v6.3.7 (31d656c0c), the release of 17 October 2026, has them, each keeping what this wiki adds of its own, and each header says so and what 6.3.7 changes in it, where the nine brought up on 17 October 2026 named the batch 6.3.7 was to be made from, at the same commit, and said that the rest of the copy was v6.3.6's. Re-copied from the tag: Security Policy (SECURITY.md: its latest release 6.3.7, published on 17 October 2026; "6.3.7 was the exception" under What gets a release, and step 3's pointer to it; Every change and Declaring a finding not exploitable; the release's scans as the second evaluation; the libraries dependency-scan says UNKNOWN for), Release Process (RELEASE.md: the release's notes and their Download table, step 9's gate asked before the tag with master's copy of the exceptions, 6.3.7's paragraph under A security release during the pause), Contributing (CONTRIBUTING.md: the runners each pipeline runs on, the change checks, the eighteen fetched inputs, the new checkers and generated files, the secret scans, sign-off since batch41, step 6 of Merge requests), Governance, Assurance Case, Regression Test Environment (Connector/C 3.4.11's folder; the 7-Zip path the document now writes right), The Service Account (/ProtectFolders hardening the program folder, Bin and ControlPanel, and the steps before upgrading an installation at a drive's root), The Update Feed (the Control Panel's Updates page, the summary past the Download table, the pages job on the project's own runners, where this copy said GitLab's shared runners), Upgrading Guide (the route between backends, the .NET 10 runtime setup leaves as it is), Third-Party Binaries (MariaDB Connector/C 3.4.11 built from MariaDB's source and fetched from build-inputs-2), Database TLS, Diagnosing Stalled Mail (the spam tests' failed lookups and the DNSSEC deferral's reasons), Event Scripts, Fuzzing (UTS 46 against ada), High Availability Runbook (the PROXY protocol behind a load balancer), Secrets at Rest, Secure Defaults, The SNMP Agent, Migrating the Database Backend, Regulatory Scope and Architecture (a transaction is its thread's). Command Line is unchanged at the tag; its header says so and its links point at it. A link to a document the wiki copies is a wiki link; every other link to a repository file that pointed at v6.3.6 points at it as v6.3.7 has it. The headers no longer list as this copy's departures what the tag's documents now say themselves: the day 6.3.6 was published (Security Policy, Contributing), the count of the findings accepted at 6.3.4 (Security Policy), the report of a command run as root on Linux (The Service Account), the route between backends (Upgrading Guide) and the 7-Zip path (Regression Test Environment). Release Process's notes marked Now at steps 9 and 11 say that every release since 6.3.5 has been made by its tag's pipeline, where step 9's said that no pipeline had run on a release tag, and that the installer's inputs are fetched from the build-inputs-1 and build-inputs-2 releases. Signed-off-by: Christopher Holloway <chris.holloway@progressiverobot.com>

Christopher Holloway committed Oct 18, 2026

The twenty-two pages that copy a repository document are copied again from the documents as the tag v6.3.6, the release, has them, each keeping what this wiki adds of its own. Re-copied from the tag v6.3.6 (e3c448aa6), each header saying so and what 6.3.6 changes in it: The Update Feed (UpdateFeed.md), The Service Account (ServiceAccount.md), Security Policy (SECURITY.md), Release Process (RELEASE.md), Upgrading Guide (Upgrading.md), Secure Defaults (SecureDefaults.md), Third-Party Binaries (ThirdPartyBinaries.md), Contributing (CONTRIBUTING.md), Secrets at Rest (SecretsAtRest.md), Assurance Case (ASSURANCE-CASE.md), Event Scripts (EventScripts.md), Command Line (CommandLine.md), Governance (GOVERNANCE.md), Regression Test Environment (RegressionEnvironment.md), The SNMP Agent (Snmp.md), High Availability Runbook (HighAvailabilityRunbook.md), Regulatory Scope (RegulatoryScope.md) and Fuzzing (Fuzzing.md). Architecture (ARCHITECTURE.md), Database TLS (DatabaseTls.md), Diagnosing Stalled Mail (DiagnosingStalledMail.md) and Migrating the Database Backend (MigratingDatabaseBackend.md) are unchanged at the tag; their headers say so and their links point at it. The wiki's own additions - the diagrams, the tables and the notes each header names as the wiki's - are kept. A link to a document the wiki copies is a wiki link; every other link to a repository file points at it as v6.3.6 has it. Where SECURITY.md and CONTRIBUTING.md say 6.3.6 was published on 1 October 2026, the copies give 6 October 2026, the day of its publication, and their headers say so. Where Upgrading.md says there is no first-class route from one database backend to another, the copy keeps its pointer at Migrating the Database Backend, the repository's own document on it. Secrets at Rest, "Who can open it": a wiki note under the paragraph that still describes the permissions hMailServer.ini inherits under Program Files, which 6.3.6's /ProtectFolders closes. Contributing, the wiki's own Service Desk section: the address at the group's path, as SECURITY.md gives it. Signed-off-by: Christopher Holloway <chris.holloway@progressiverobot.com>

Christopher Holloway committed Oct 5, 2026

Every link to the project points at its address in the group, gitlab.com/hmailserver/hmailserver, where it has been since 29 September 2026; the identities 6.3.5 was signed with, and the setting row that lists both paths, keep the path the project began at. The project's path in all 63 pages that linked it (548 links), 6.3.6 being the first release made at the group's path. Left as they were: the seven signing identities https://gitlab.com/Progressiverobot/hmailserver//.gitlab-ci.yml@refs/tags/v... (Release Process, Security Policy, The Update Feed, and the UpdateSigningIdentity row of Settings Reference), which name what 6.3.5 was signed as; and the UpdateSourceRepository row of Settings Reference, which lists both paths on purpose. The container registry's name is corrected with the container page. Signed-off-by: Christopher Holloway <chris.holloway@progressiverobot.com>

Christopher Holloway committed Oct 5, 2026

Seven features 6.3.4 added have a page of their own, each a copy of its repository document as the tag v6.3.4 (51e2dbda0) has it, headed by the Source of truth note the other copies carry (GitLab issue #25): The SNMP Agent (Snmp.md), Event Scripts (EventScripts.md: the JavaScript engine of the Linux builds, and the contract for all three languages), Secure Defaults (SecureDefaults.md), The Service Account (ServiceAccount.md), Secrets at Rest (SecretsAtRest.md), Database TLS (DatabaseTls.md) and The Update Feed (UpdateFeed.md). A link to a document the wiki copies became a wiki link - Command Line, Third-Party Binaries, Warm Standby, High Availability Runbook, Upgrading Guide and The Service Account, the last two to their sections - and every other became a link to the document at v6.3.4. Each note says where the tag has moved on from its document. The installer smoke test SecureDefaults.md expects on a branch is in the tree at the tag, and does not yet make the checks it lists. ServiceAccountName and ServiceAccountPassword, which ServiceAccount.md calls no longer read, are still loaded by the settings loader (IniFileSettings.cpp:457-458) and used by nothing. The site that UpdateFeed.md's opening note calls not yet deployed has served the feed over HTTPS since 26 September 2026, and at 05:31 UTC on 27 September answered /releases/latest and /releases.json with 6.3.4 and /download/hMailServer-x64.exe with a 302 to 6.3.4's installer. The sidebar and the landing page list each page in its group: Database TLS, Secure Defaults, The Service Account and Secrets at Rest under Secure it; The SNMP Agent and The Update Feed under Operate it; Event Scripts under Extend it. The landing page counts 72 pages (the 74 files less the sidebar and the footer), the other 71, and twenty-two with a Source of truth note; the guides stay 38, since a copy is a repository document. The Runbooks Digest names the seven among the copies. Monitoring and Health, The COM API and Scripting, Security Hardening, Encryption and Certificates, Migrating the Database Backend and Upgrading hMailServer keep what they say of each feature, and gain a line pointing to its page. Every page was rendered with GitLab's Markdown API in the project's context on 27 September 2026: every wiki link, relative link, anchor and link into the repository, at master and at v6.3.4, resolves, and nothing on the new pages is HTML that could be stripped. The one finding, a link to Sigstore's cosign on GitHub from Installing hMailServer, is that project's own page and predates this change.

Christopher Holloway committed Sep 27, 2026

The wiki is brought up to the published release: 6.3.4 is the signed tag v6.3.4 at 51e2dbda0, published on GitLab on 27 September 2026 (02:37:56 UTC), not 8adce850a of 26 September, where the tag was first made and whose release gate failed. Every page that named 8adce850a as the tag or the release names 51e2dbda0, and every page that dated the release 26 September dates it 27 September: the footer, the sidebar, the page headers, the landing page's Current release, the release timeline's row and picture, section 4e's heading. Each fact those pages took from 8adce850a was checked at 51e2dbda0: the commits between the two change 42 files, whose only documents are CHANGELOG.md, BrandLogos.md, OpenPGP.md and SecretsAtRest.md, none of which a page copies, and which include none of Constants.h, Version.h, formMain.cs, the create scripts, IniFileSettings.cpp, SettingsDefaults.h, the IDL, the Control Panel or .gitlab/allowed_signers, so every file and line cited at the tag holds. Where a page said it read the tag on 26 September, it says it read 8adce850a and that what it read is the same at the tag. The suite has 3,811 [Test] attributes at the tag, not 3,809 (the fixes add two tests); twelve places say so. Changes Since 6.2.10 gains, in batch 40, what the release gate found and the eleven fixes: three change tests alone; of the other eight, three fix what 6.3.3 did too - a relay that refused the OAuth2 login got the server's token into the bounce and the logs (open since 6.2.22-pre1), a rule's forward, copy, reply or notice and an account's forwarding could wait a minute in the queue, and IMAP LIST and LSUB took a literal's length for the pattern - and five fix code new in 6.3.4 (sealed mail and an unnamed part, a closed key's FETCH refusal logged as a critical error, the help desk's password reset read as bytes, the logo and key-directory holds, an alert's encoded Subject over 76 characters). The gate table and the timeline give the re-run on 51e2dbda0 as the logs have it: 2,349 on the first bench (2,252 passed, 0 failed, 97 skipped) and 1,462 on the second (1,446 passed, 0 failed, 16 skipped), the 71 live-update tests elevated, all passed, and the Linux suite on the packaged binary, 2,013 tests, 1,562 passed, 0 failed, 451 skipped (results/b40full3 on linix); the database upgrade gates, the security scans, the fuzzers and the installer smoke test, 65 of 65. The JScript flaw is CVE-2026-100741 (CVSS 3.1 9.8), and the project's issue #43 is public: Security Hardening 17.11, the scripting page, the changes, the capabilities and the API reference name them. The project's download links work (the site has served HTTPS since 26 September; on 27 September they led to 6.3.4's files), so Installing on Linux and Running in a Container no longer say they work once it does. The copies of SECURITY.md and CONTRIBUTING.md give the day GitLab published the release, and the security policy's copy counts the 455 findings .gitlab/security-exceptions.yml accepts at the tag - the 169 of the SAST triage and the 286 that held the release scan - where the repository document still says 169; each header says where the copy differs. Three places said no pipeline had run on the fixed tree; one has, the schedule of 27 September 2026 on master 6bc09509d, which ran the update feed's pages and package-repos jobs alone, and they say so.

Christopher Holloway committed Sep 27, 2026

The wiki is brought up to the release itself, the signed tag v6.3.4 at 8adce850a: what it said of 6.3.4 had been read from master 92b8f9f63, which the tag's last batch, batch40, follows. Schema 6104, not 6103: 6.3.4 is sixty-four steps from 6.3.3's 6040, the last, 6103 to 6104, adding hm_fetchaccounts.faorigin (Constants.h:1224, formMain.cs:314 and the three create scripts at the tag); every header, table, count and diagram that gave 6103 as 6.3.4's schema says 6104, the upgrade table's step counts are each one more, the step tables gain the 6104 row, and where 6103 means the secrets-at-rest step it stays. No page calls 92b8f9f63 the release or the tree 6.3.4 was cut from any more: a page checked against it says so and that the tag adds a last batch, and a page that copies a repository document the tag leaves as it was says the document is the same at the tag. The copies of repository documents batch40 changes are brought up to the tag, hunk by hunk from the version each copy had reached, their own additions kept: Security Policy (the SSH-signed sums of a release made by hand and the two commands that check a download, an SBOM for every compiled file, a release held by its security scans, secrets refused before they are pushed; its note no longer says the document calls 6.3.3 the latest), Release Process (A release made by hand, the security gate, the SBOM jobs), Upgrading Guide (the installer's HM_ADMIN_PASSWORD, the named lock on MySQL and MariaDB, the probe that stops an upgrade on too old a server at 6041 and the repair of a half-done one), Contributing, Governance (rebuilt on the tag's document, which gives the GitLab roles the owner set on 26 September 2026; the map and the note on code-signing material kept), Assurance Case, Regulatory Scope, Third-Party Binaries, Fuzzing, Migrating the Database Backend and Regression Test Environment. The installer's password handling as the tag has it: the password page is skipped for HM_ADMIN_PASSWORD as for /adminpassword, the database step hands the password (or the live update's token, as token:<hex>) to DBSetupQuick and DBSetup in HM_ADMIN_PASSWORD and DBSetupQuick to DBUpdater the same way, never on a command line, the switch table has the variable, and the unattended upgrade takes either (Installing hMailServer, Upgrading hMailServer 18.4a). How to verify a download: 6.3.4's hmailserver-6.3.4-SHA256SUMS.txt and its .sig, checked with ssh-keygen -Y verify against .gitlab/allowed_signers and sha256sum -c, as SECURITY.md at the tag gives them, and each installer's and package's own SBOM (Installing hMailServer 4.1, Installing on Linux, Upgrading hMailServer 18.4 and 18.4c, home, Release Process, Project Handbook, Troubleshooting, Glossary, Assurance Case, Regulatory Scope). The /download/ links no longer wait on issue #30: the feed's site has had its certificate since 26 September 2026. Changes Since 6.2.10 gains batch 40 in section 4e (the audit before the release, GitLab's SAST and the release gate, the installer's password, the SBOMs and signed sums, the stable download links, the release chain and landing through merge requests, what the first pipeline found, Windows on ARM compiled), items 47 to 51 in section 5 (reserved domain names, a one-time code taken once, the outbound relay's OAuth2 token over TLS only, the fetchers and schema 6104, a pre-6.3.4 hash of a password above U+00FF refused), SQLOLEDB in item 38, the 6041 probe in item 36, and the installer's own PBKDF2 in item 39. Also what the tag makes false elsewhere: the suite has 3,809 [Test] attributes, not 3,698; hmailserver/docs holds 105 documents; Windows on ARM compiles and has never run, where it was "not built"; IMAP offers STARTTLS only before sign-in and TLS and COMPRESS=DEFLATE only after sign-in (Capabilities and Configuration); a parent-domain passkey relying party needs passkey_allowed_origins (Sign-in and Identity); the Settings Reference counts 193 hm_settings rows with PasskeyAllowedOrigins and WebPushLoopbackHttpForTests; and the Linux suite's CI job first ran in the pipeline of 26 September 2026 rather than never.

Christopher Holloway committed Sep 26, 2026

The eighteen pages that copy repository documents are copied again from master 92b8f9f63, the 6.3.4 release, so they say what the repository says on the day 6.3.4 is published. Each was brought forward from the version it was last copied from, 540347500, hunk by hunk, and the wiki's own additions were kept where they are still true. What the documents gained: the stored-secrets key in the command line, the high-availability runbook and the warm standby, which now say that it is made once, exported and imported on the other node, and what a node without it does (HM7495); the service account, the COM rule that a change to an item of a shared collection is written onto a copy until Save(), and MySQL's utf8mb4 step in the upgrading guide; database TLS and utf8mb4 in the backend migration; the fuzz job moved from GitHub Actions to GitLab CI's linux-fuzz; the Windows and Linux suites on GitLab CI in the regression environment; the security policy's scans on the project's own runner and its questionnaire pointer; the contributing guide's CI checks; MessageStore and QuickJS in the architecture; a pre-release tag's suffix and the pinned .NET runtime in the release checklist; the update verifier's two signers in the assurance case. Third-Party Binaries follows a rewritten document: the tree commits no executable binary, and the sections this copy had added about where the forty committed binaries lived and their dispositions described the tree before 11 September 2026, so they are gone. Regulatory Scope's note loses four of the five things it called out of date, which the repository has corrected, and its own Annex I table now says the installer has been Authenticode-signed since 6.3.1, that 6.3.x is the supported line, and what a new installation's secure defaults are. The wiki's labels that said "in the tree after 6.3.3, not yet released" now say what is new in 6.3.4, and each header names 92b8f9f63. Where the repository text itself is now stale - the release checklist's and Regulatory Scope's statements that no release can be cut while the chain is on GitHub, the contributing guide and the security policy naming 6.3.3 as the latest release, the assurance case pointing at the changelog's Unreleased section - the copy keeps the repository's words and the header says so where a reader would be misled; 6.3.4 was built and Authenticode-signed by hand and carries no Sigstore bundle.

Christopher Holloway committed Sep 26, 2026

The pages that copy repository documents are copied again from master 540347500, so they say what the repository says after the move to GitLab and after the batches that have landed since each was last copied. Contributing and Security Policy follow their files to the root of the repository, where both were rewritten for GitLab: the contributing guide is now the full one - where things are, how to report, the rule for GitHub's issue numbers, the prerequisites, one build command per component, the suites, every checker, the change checklists, commits, sign-off and merge requests - and the security policy supports 6.3.x and takes a report by a GitLab confidential issue or the project's Service Desk address. Governance names two maintainers, and it, the Release Process and the Assurance Case carry the repository's note that the release chain runs on GitHub Actions and cannot run until it moves. The runbooks gain what the tree after 6.3.3 added to them: the settings store in the database and a save that applies the setting, the drain as a state of its own and UpgradeOnStart, the message held for a sandbox, SPF as an RFC 7208 implementation, the Linux update helper, the locked copies of shared collections, the second bench and the gate in halves. Command Line becomes the whole current document, with every verb the clients have gained since 15 September. Most of these pages carry the wiki's own additions - diagrams, tables, worked commands, and corrections the repository has not taken - so each was merged three ways, from the version of the document it was last copied from to master, and the additions were kept wherever they are still true. Where the repository text describes something in the tree after 6.3.3 and not yet released, the page labels it, as the wiki does elsewhere, and says what 6.3.3 does instead where that matters to someone running it. Additions the repository has overtaken are gone or rewritten: the contribution flowchart and the table of required checks, which were GitHub's, are gone; so are the governance duty table and the bus-factor section, which the document now covers with two maintainers; the security flowchart and timeline send a report to GitLab and ask for the CVE there. The Project Handbook and the release diagram describe build/generate-com-wrapper.ps1, which replaced build/regenerate-interop.ps1 on 11 September 2026 when the COM wrapper left git. Each copy says at the top that it was re-copied from master 540347500 on 24 September 2026. Two say plainly where the repository document is itself out of date, and the copy with it: Third-Party Binaries and Regulatory Scope still count forty committed binaries, none of which has been in git since 11 September 2026, and Regulatory Scope still calls Authenticode signing outstanding and the releases GitHub's. The regression environment's 7-Zip path, which the repository document writes with a control character where \7 belongs, reads correctly here. Fuzzing.md has not changed since the wiki's copy, and its page says so.

Christopher Holloway committed Sep 24, 2026

The wiki's links move from GitHub to GitLab, where the project has lived since 22 September 2026. GitHub suspended the account on 18 September and every one of the project's GitHub addresses answers 404, so the download links, the source links, the security route, the support route and the "report an error" link all led nowhere. Every link to a file in the repository now opens it on GitLab, and the four community files that moved to the repository root on master 540347500 - CONTRIBUTING.md, SECURITY.md, SUPPORT.md and CODE_OF_CONDUCT.md - are linked there, which also mends the seven links to a root SECURITY.md that did not exist when they were written. The download links open the latest GitLab release, and the installer has a link that always fetches the latest one. A private report goes to a GitLab confidential issue or the project's Service Desk address, a question is an issue with the Question template because GitLab has no Discussions, and a reader who wants to hear of releases is told how to watch the project on GitLab. The Project Handbook, written against GitHub on 4 September, says that its GitHub process is the project as it was then. The GitHub facts that stay are facts about the product, and each now says what is true of it: the update feed the server reads by default, which answers 404, so a check fails with "The update feed answered HTTP 404." until UpdateFeedUrl names another feed; the Sigstore identity and issuer that signed every release up to 6.3.3, which verify those files wherever they are hosted and which the verifier's defaults still name, so a release signed anywhere else is refused unless the three settings are changed; and the two GitHub addresses compiled into the product, the Policy line of the security.txt the server serves and the Control Panel's About link. Every GitHub issue, discussion and pull-request number written with a hash is now written in words - "GitHub issue 114", "GitHub discussion 18", "upstream PR 602" - with the kind the repository's CHANGELOG gives each number, because GitLab turns a hash and a number into a link to its own issue of that number, which is something else. The one maintainer handle left in running text is in code, and the webmail's mentions no longer read as a GitLab user. Four things GitLab rendered wrongly are put right. The Stopping Spam link to Routes and Relays was split by a line break and led to a page that does not exist. The Fuzzing page's two repository-relative links resolved inside the wiki; they open the fuzz folder and its regression README on GitLab. Placeholders written outside code, such as <host:port>, <addr> and <culture>, were read as HTML and dropped or turned into links; they are in code now. And the Assurance Case's eight section headings began "1. " above a setext rule, which CommonMark reads as a list and a thematic break, so its contents links had nothing to land on; they are ATX headings with the ids those links name.

Christopher Holloway committed Sep 24, 2026

The wiki describes the tree after 6.3.3 as it lands on master: the settings store, the audit trail and alerts, what a domain does to a message, remote-domain policy, a backup that carries everything, and the rest of three batches Twenty-seven pages, each written from the code at 3233b0cc4 and then checked sentence by sentence against it by a second reader: 166 claims verified, 23 corrected before this commit. Everything the tree holds that no release has is labelled as in the tree after 6.3.3 and not yet released, and the next release is called that rather than given a number. The changes page carries the four schema steps 6042 to 6045 in section 2, the second and third batches in section 4e beside the first, the gate for 3233b0cc4 (2,410 tests, 2,402 passed, none failed, 8 skipped; the Linux figures are a placeholder until that run lands), and in section 5 the changes an upgrade makes with no switch - above all that the database is the settings store, that an edit to [Settings] in hMailServer.ini is undone at the next start and reported as HM5804, what to do instead, and that the password pepper stays in the file and out of every backup, so a restore onto new hardware needs it brought along. Fourteen pages that told a reader to edit the file now say how the same setting is changed once the store lands, and what an edit to the file does then; the settings reference says the table of INI keys is closed. The feature pages cover the backup that carries every per-account store, the audit trail and alerts with what each does not yet record or raise, syslog export, per-domain reports, remote-domain policy and mandatory TLS, SPF as RFC 7208, hmctl's configuration document, the tours, and the REST routes counted at the commit rather than guessed at. The checking found things the tree itself gets wrong, which are recorded for the next batch rather than papered over here: hmctl's configuration document has an empty ini section because it reads a route that answers names only; the packaged Linux ini's RestApiPort=0 is adopted into the store at first start, so the packaging README's edit-the-file way of turning the REST listener on no longer works; the OpenAPI text for the ini routes and the note the server writes into hMailServer.ini still describe the file as the store; and five repository documents say what their wiki copies said before this commit.

chrisholloway5 committed Sep 15, 2026

The wiki rewritten in depth: 141 diagrams drawn from the code, every reference page made exhaustive, and twenty-one defaults corrected that no page had ever stated The wiki was accurate after this morning's sweep and thin. This is the depth pass the owner asked for: every page read against the code again, every reference page completed, and a diagram wherever a diagram says it better than a paragraph. 141 Mermaid diagrams across 49 pages, each drawn from the handler rather than from memory: the SMTP submission sequence with the checks in the order ProtocolMAIL_ and ProtocolRCPT_ actually run them, the IMAP session state machine from the per-command authentication guards, the DMARC evaluation, the delivery pipeline and its retry arithmetic, the ACME renewal state machine, the spam decision from connection to delivery, the schema chain, the component and threading models of the whole server, the CI workflows and what each one gates, the release draft-sign-publish flow, and a map of the journeys through the wiki itself on the landing page. The reference pages are now exhaustive rather than representative: every route with its method, authentication, request and response shape and its error codes; every [Settings] key with its default, its effect and whether it needs a restart; every port with the key that opens it; every rule criterion and action; every Sieve extension the server advertises; every metric the listener exposes. Twenty-one facts were wrong or absent, and each is corrected against the code rather than adjusted. The ones that will save somebody an afternoon: IMAP SASL PLAIN and SCRAM are not advertised on a default install, which looks exactly like a client bug; SMTP AUTH PLAIN is off by default too; DMARC checking is on while SPF and DKIM as scored tests are off, and the DMARC test evaluates both itself regardless; the seeded DNSBL and SURBL servers are inactive; a new mailbox contains INBOX and nothing else, which is the real cause of the classic missing Sent folder; MTA-STS hosting and autoconfiguration default on while both web-service ports default to 0, so the shipped default is features on with nothing listening; and TLS-RPT and DMARC report sending are inert until their From address is set, with each completed day discarded unsent. Eleven screenshots are named as wanted, with what each should show. The existing images are from a 6.2.10-era build and the pages say so where it matters.

chrisholloway5 committed Sep 8, 2026

Every page read against the tree on 8 September 2026: the release and schema stamps (6.2.27, schema 6031, 2,124 tests), the routes, keys and pages that were missing, the listener paragraphs that were false, and the unreleased work labelled as such An audit of all 51 pages against master produced about 360 findings on 43 of them, applied here. The largest: the APIs reference lacked 47 routes and described the listeners as single-threaded HTTP/1.0 (they run on the Boost.Asio HTTP/1.1 server); the settings reference lacked the fourteen Update* keys and IMAPCompressionEnabled and counted 195 keys instead of 236; every "current release" stamp said 6.2.24; Changes Since 6.2.10 ended three releases and one schema step early; Architecture said the optional listeners were deliberately not Boost.Asio and built their own SSL contexts (neither is true); the upgrading pages still warned of a silent-upgrade hang fixed in 6.2.23; the Control Panel pages knew nothing of the seventeen languages, the Updates card or the stalled-mail page; and two pages did not render (a header inside a code block, CR and TAB bytes in file names). Anything in the tree but not in v6.2.27 - the live update, the self-service portal and /api/v1/me, the HttpServer listeners, IMAP COMPRESS=DEFLATE, the seventeen-language Control Panel, the Control Deck views - is documented and marked "in the tree after 6.2.27, not yet released". The Control Panel screenshots are still from a 6.2.10 build and the pages now say so.

chrisholloway5 committed Sep 8, 2026

References regenerated for 5 September 2026: message retention, scrypt, the ClamAV connection test, tarpitting, the server-enforced administrator second factor, and the assertion build; the landing page keeps its new name

chrisholloway5 committed Sep 5, 2026

Settings and Control Panel references: the five keys added on 5 September 2026 (special-use folders, MariaDB TLS opt-out, hash work factors), and the migration guide's MySQL row

chrisholloway5 committed Sep 5, 2026

The hMailServer 6.2.24 wiki: the corrected manual, the runbooks, and the references generated from the code

chrisholloway5 committed Sep 4, 2026