V4 - Firebird ODBC Driver Modernization #275
Conversation
Remove obsolete infrastructure accumulated over 20+ years: Build systems (Builds/): - 14 platform-specific configurations: MSVC 6.0-2022, MinGW, Borland C++, GCC for Linux/macOS/FreeBSD/Solaris/AIX/HP-UX - Replaced by CMake (added in following commit) Vendored headers: - FBClient.Headers/: 60 Firebird header files (~25K lines total, including IdlFbInterfaces.h). Now fetched via CMake FetchContent at build time - Headers/: Microsoft ODBC SDK (SQL.H, SQLEXT.H). System SDK headers used Removed components: - OdbcJdbcSetup/: GUI setup dialog (30 files) — use odbcconf/odbcinst - Install/: HTML help (5 languages), InnoSetup scripts — replaced by WiX - Test/ + Tests/ + JdbcTest/ + TestInstall/: old test infra — replaced by Google Test suite - Res/: locale resource files (5 languages) — i18n not supported Dead code from driver source: - LinkedList.cpp/.h: replaced by std::vector (Phase 5) - Lock.cpp/.h: unused RAII wrapper - ServiceManager.cpp/.h: backup/restore GUI support (880 lines, no callers) - SupportFunctions.cpp/.h: ODBC escape sequences (removed by design) - Engine.h: obsolete macros conflicting with <algorithm> - IscOdbcStatement.cpp: merged into IscStatement - LoadFbClientDll.h: replaced by new version in src/IscDbc/ - OdbcDateTime.cpp/.h: replaced by FbDateConvert.h - TransactionResourceAsync.cpp/.h: ATL/DTC support (removed) - WriteBuildNo.h: replaced by git-based versioning - Build artifacts: .exp, .manifest, MinGW .def Old CI: msbuild.yml, linux.yml, arm64 variants ChangeLog files: ~4,600 lines (git history replaces these)
…ments This commit contains all driver source code changes, the new CMake build system, and CI/CD infrastructure. Source files are moved from the root into src/ and src/IscDbc/ subdirectories. BUILD SYSTEM: - CMake 3.20+ replaces all legacy build systems - Firebird headers fetched via FetchContent (pinned to v5.0.2) - Google Test/Benchmark fetched via FetchContent for testing - LTO enabled for Release builds; SRWLOCK on Windows - WiX v5 MSI installer (installer/Product.wxs) - Git-based versioning (cmake/GetVersionFromGit.cmake) - Build script: firebird-odbc-driver.build.ps1 (Invoke-Build) - CI: GitHub Actions for Windows x64 + Linux x64/ARM64 CRASH FIXES (Phase 0): - GUARD_HDESC/HSTMT/HDBC macros: null check before dereference - All ODBC entry points: null handle checks (SQL_INVALID_HANDLE) - postError: snprintf with 512-byte buffer (was sprintf into 256) - 64 unsafe C-style exception casts replaced with catch clauses ODBC SPEC COMPLIANCE (Phase 1): - ISC→SQLSTATE mapping: 121 states, 100+ ISC codes, 130+ SQL codes - Dual ODBC 2.x/3.x SQLSTATE support (getVersionedSqlState) - SQLGetDiagRec: returns SQL_NO_DATA (not SQL_NO_DATA_FOUND) - SQLGetDiagField: null StringLengthPtr check - SQLGetInfo: fixed numeric storage and NULL InfoValuePtr handling - SQLCloseCursor: returns 24000 when no cursor open - SQLDescribeColW: returns SQL_WCHAR/SQL_WVARCHAR (not SQL_CHAR) - SQLSetConnectAttr: default error path for unknown attributes - SQLSetStmtAttr: cursor-state validations (24000/HY011) - BufferLength validation for W APIs (HY090 for odd values) ENTRY POINT HARDENING (Phase 2): - try/catch on 9 methods missing exception handling - clearErrors() at all entry points - Thread safety always enabled (removed DRIVER_LOCKED_LEVEL_NONE) FEATURES (Phases 4, 8, 9, 11): - Savepoint isolation: per-statement savepoint/rollback in autoCommit=OFF - Server version detection: getServerMajorVersion/Minor for FB3/4/5 - ConnSettings: SQL executed on connect via connection string param - Array parameter binding: column-wise and row-wise, operation ptr - IBatch: single server roundtrip for batch params on FB4+ (with inline BLOB support via BLOB_ID_ENGINE policy) - SQL_GUID: mapped from CHAR(16) OCTETS / BINARY(16), with all conversion methods (GUID↔string, GUID↔binary) - ODBC 3.8: SQL_OV_ODBC3_80, SQL_ATTR_RESET_CONNECTION, driver version "03.80", SQL_GD_OUTPUT_PARAMS, SQL_ASYNC_DBC_FUNCTIONS - SQLGetTypeInfo: thread-safe (per-instance copy), sorted by DATA_TYPE, multi-row returns, FB4+ type dedup - Query timeout: timer-based cancelOperation(fb_cancel_raise), SQLSTATE HYT00 on expiry - SQLCancel: functional via cancelOperation - Connection pool reset: rollback pending TX, close cursors, reset statement attributes OO API MODERNIZATION (Phase 9): - ~35 dead ISC function pointers removed from CFbDll - IscUserEvents migrated from isc_que_events to IAttachment::queEvents - TPB construction via IXpbBuilder (replaces manual byte-stuffing) - Julian-day date math replaced by shared FbDateConvert.h helpers - Error handling unified: both OO and legacy use IUtil::formatStatus - isc_vax_integer replaced by inline helper - Concrete IscDbc classes marked final (devirtualization) PERFORMANCE (Phase 10): - SRWLOCK replaces Win32 kernel Mutex (~20ns vs ~1-2μs) - 64-row prefetch buffer in IscResultSet (amortizes fetch overhead) - ConvertingString: 512-byte stack buffer (eliminates heap allocs) - BLOB object pooling per result set - Value::getString buffer reuse - clearErrors fast path with [[likely]] - std::to_chars for float→string conversion - Sqlda metadata rebuild skip on re-execute - ODBC_FORCEINLINE on hot conversion helpers - alignas(64) on Sqlda::buffer UNICODE (Phase 12): - ODBC_SQLWCHAR typedef (always 16-bit on all platforms) - All codec signatures use ODBC_SQLWCHAR* (not wchar_t*) - *ToStringW functions use SQLWCHAR* directly (fixes Linux) - Single UTF-8↔UTF-16 codec (Utf16Convert.h) - OdbcString class: UTF-16-native metadata caching - Direct W-API output: SQLDescribeColW, SQLColAttributeW, SQLGetDescFieldW, SQLGetDescRecW, SQLGetDiagRecW, SQLGetDiagFieldW bypass ConvertingString entirely (zero-copy from cached OdbcString) - CHARSET defaults to UTF8 when not specified - Linux CHARSET=NONE uses UTF-8 codec (not locale-dependent) CODE QUALITY (Phases 5, 13): - OdbcError: std::vector<unique_ptr> replaces raw linked list - IscConnection/Statement/ResultSet: std::vector replaces LinkedList - OdbcObject/OdbcError/OdbcEnv: private/protected visibility - EnvShare: Meyer's Singleton (construct-on-first-use) - returnStringInfo: SQLINTEGER* delegates to SQLSMALLINT* overload - snprintf/swprintf macro guards for MSVC >= 1900 - .clang-format configuration - i18n code removed from ConnectDialog.cpp
New test suite using Google Test, replacing the old MSBuild/VSTest-based tests. Tests cover the full ODBC API surface against a real Firebird 5.0 database, with graceful GTEST_SKIP() when no connection is available. Test organization (by topic): - test_null_handles.cpp: 65 tests via direct DLL loading (no Driver Manager) — every ODBC function with NULL/invalid handles - test_connect_options.cpp: connection attributes, timeout, async mode, query timeout, connection reset (36 tests) - test_catalogfunctions.cpp: all 12 catalog functions + type info ordering and version-gated types (29 tests) - test_result_conversions.cpp: data type conversions in results — INT, BIGINT, FLOAT, DOUBLE, NUMERIC, VARCHAR, DATE, TIME, TIMESTAMP, truncation, NULL, cross-type (35 tests) - test_param_conversions.cpp: parameter type conversions (18 tests) - test_data_types.cpp: SMALLINT through TIMESTAMP, precision, scale, parameterized insert/select (18 tests) - test_array_binding.cpp: ODBC Array of Parameter Values — column-wise, row-wise, NULL handling, operation ptr, 1000-row, multi-type (17 tests) - test_errors.cpp: error handling, SQLSTATE mapping, diagnostic row count, truncation indicators (18 tests) - test_descriptor.cpp: IRD/ARD/IPD, SQLGetDescRec, SQLCopyDesc, desc field count, copy-desc crash fix (13 tests) - test_odbc38_compliance.cpp: ODBC 3.8 features — env version, driver version, getdata extensions, async DBC, reset connection (12 tests) - test_guid_and_binary.cpp: SQL_GUID mapping, UUID roundtrip, type coverage, FB4+ types (14 tests) - test_prepare.cpp: SQLPrepare/Execute lifecycle (10 tests) - test_cursors.cpp: cursor behavior, commit/rollback, close/re-exec (7) - test_cursor_commit.cpp: cursor behavior across transactions (6 tests) - test_cursor_name.cpp: set/get cursor name, truncation, duplicates (9) - test_scrollable_cursor.cpp: all fetch orientations (9 tests) - test_stmthandles.cpp: 100+ simultaneous handles, interleaving (4) - test_multi_statement.cpp: multi-handle on one connection (4 tests) - test_blob.cpp: small/large/null BLOB read/write (3 tests) - test_savepoint.cpp: savepoint isolation across errors (4 tests) - test_escape_sequences.cpp: passthrough verification (6 tests) - test_data_at_execution.cpp: SQL_DATA_AT_EXEC / SQLPutData (6 tests) - test_bindcol.cpp: dynamic unbind/rebind mid-fetch (5 tests) - test_wchar.cpp: SQL_C_WCHAR bind/fetch, truncation, empty (8 tests) - test_descrec.cpp: SQLGetDescRec for all column types (10 tests) - test_conn_settings.cpp: ConnSettings connection param (3 tests) - test_server_version.cpp: server version detection (4 tests) - test_odbc_string.cpp: OdbcString UTF-16 class unit tests (26 tests) - bench_fetch.cpp: Google Benchmark for fetch throughput Infrastructure: - test_main.cpp: shared GTest main with connection string from env - test_helpers.h: FirebirdODBCTest fixture, SQLWCHAR comparison helpers - tests/CMakeLists.txt: GTest + GBenchmark via FetchContent
- Docs/FIREBIRD_ODBC_MASTER_PLAN.md: comprehensive registry of all known issues, architectural analysis, improvement roadmap, and implementation status. Serves as the single source of truth. - Docs/firebird-api.MD: Firebird OO API reference for driver authors - Docs/firebird-driver-feature-map.md: feature comparison matrix - Docs/CONVERSION_MATRIX.md: OdbcConvert dispatch table documentation - Docs/PERFORMANCE_RESULTS.md: benchmark methodology and results - AGENTS.md: guidelines for AI-assisted development on this project
|
Oh my Lord... |
Isn't support of those escapes required for ODBC compliance? Or did ODBC do away with them? |
Agree with @mrotteveel What's the reason to drop this logic? I remember fixing an issue with these escapes a while back - #207. It was 2.5 years ago only, so we can suggest this function is on demand and is being used. |
This shouldn't be necessary. As far as I know, the engine already uses server-side savepoints to undo work on statement failure. A statement failure does not "corrupt" a transaction (and if it does, that would be a bug that should be fixed in the engine, not in a client). If execution of an individual statement fails, all its effects are already undone, so trying to do the same client-side is just unnecessary overhead. Using savepoints client-side only makes sense when covering multiple executions. |
I'm not so familiar with the ODBC landscape, but is there a specific reason for this (e.g. is this deprecated technology?). Firebird supports real two-phase commits, and thus support for distributed transaction should be available, I think. |
Isn't this the thing that is shown in the ODBC Data Source Administrator in Windows? If so, it should remain in place. |
+1! Naturally it is! |
|
Thanks for the feedback, guys! I will address them soon. 👍🏻
Well... There's always room for improvement 😄 |
There's actually a choice, one that Microsoft calls "DBMS-specific grammar": https://learn.microsoft.com/en-us/sql/odbc/reference/develop-app/choosing-an-sql-grammar
I'm using the v4-preview driver alongside a few other tools, and so far nothing has raised any issues. I removed this primarily for performance reasons -- and to make follow-on work (of reducing reallocations and string transliterations) simpler. If there’s a strong case for keeping it, we can add it back. My two cents: this is a legacy feature from the early days of ODBC, when the goal was to make applications fully database-agnostic. I took the introduction of “DBMS-specific grammar” above as a signal from Microsoft that this level of abstraction isn’t really necessary anymore. |
Mark, you’re right. I’m also finding this statement very strange. I’ll investigate it. It seems like it may be a misinterpretation of what was done when savepoint support was added. |
This was another of those "this is making things complicated so let's cut it for now" moments during the CMake migration. ATL is a very old Windows technology, and it was making it harder to keep the codebase truly multiplatform.
Another reason is that I didn't see any ATL usage in the PostgreSQL ODBC Driver (which I used as a reference). On a quick skim, it looks like they DO support distributed transactions -- so if we decide to bring them back, we shouldn’t need ATL to do it. My two cents: in modern architectures, distributed transactions are typically handled at the application level (e.g., saga patterns, event sourcing) rather than via DTC. In 25+ years of using Firebird, I particularly never once needed them. So, if it's genuinely needed, we can bring it back -- but this would increase the code complexity for the sake of benefiting only a tiny fraction of Windows users. |
You're both right, folks!
I'll dig into how we can bring that support back. Ideally in a modern way, if possible. |
Sorry guys. At first I genuinely thought @irodushka was joking. But then I realized I know basically NOTHING about ODBC drivers outside of Windows. Is there an equivalent to the ODBC Administrator on Linux or macOS?
|
|
I'm converting this PR to a draft for now. Thank you VERY much for all the precious insights, guys! I'm going back to the drawing board. For two (great) reasons:
Adopting Also: C++20 and vcpkg for dependencies! ❤️ |
That is a choice for an application developer, not an ODBC driver developer. And that section explicitly says (emphasis mine):
And
In other words, support for escape sequences is required.
They are a part of the ODBC specification, and as far as I can tell, drivers are required to support them. See above, and ODBC and the Standard CLI, so removing them is not an option. For example, SQLPrepare Function says (emphasis mine):
That is phrased unconditionally. Similar phrasing is used in SQLExecDirect Function. Maybe you haven't seen applications that use them, that doesn't mean that they're not out there. And especially given Firebird is an underdog, removing features that can bridge the gap in syntactic difference sounds like a big no to me. |
I believe unixODBC also has a configuration GUI (ODBCConfig), but I'm not familiar with it.
Looking at https://www.unixodbc.org/doc/UserManual/ it has a form, but I'm not sure how it's populated (and that page looks like it hasn't been touched since the turn of the century, with references to Star Office 5 and the GUI screenshots).
I guess (but again, I'm not that familiar with ODBC!) |
For comparison, Jaybird has over 10,000 tests. |
You are correct, Mark. It seems that in my eagerness to bring simplicity and finalize the tasks, I misinterpreted the Microsoft documentation. I'll bring them back. Ideally, with a more modern parser. 👍🏻 |
|
Please please
Do not touch the code that doesn't need to be touched) the code that works
and was fine tested about 2 years ago - see the link to the ticket above. I
mean escapes.
The best thing you can make here is creating a few additional tests around
it - that's welcome. But rewriting the code just for the sake of rewriting
it - not.
Regards
чт, 12 февр. 2026 г., 16:54 F.D.Castel ***@***.***>:
… *fdcastel* left a comment (FirebirdSQL/firebird-odbc-driver#275)
<#275 (comment)>
That is a choice for an application developer, not an ODBC driver
developer.
You are correct, Mark.
It seems that in my eagerness to bring simplicity and finalize the tasks,
I misinterpreted the Microsoft documentation.
I'll bring them back. Ideally, with a more modern parser. 👍🏻
—
Reply to this email directly, view it on GitHub
<#275 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAKZQLTEAOPGYBEUPIYJWN34LSA27AVCNFSM6AAAAACUVFLU3CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTQOJRGEYDCMJVGI>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
Sorry, what link / ticket? Also, I’ll be in dire need of those tests, now. Reintegrating the old code at this point would probably be a no-go. |
#207 as mentioned in this comment |
Thanks, Mark! My oversight. P.S. This tool would be enormously useful! UPDATE: Damn! It's a GUI! 😅 |
|
Reintegrating the old code at this point would probably be a no-go.
Please clarify this for me. Do you mean that removing this code (processing
of escape characters) cannot be undone in your PR?
чт, 12 февр. 2026 г., 19:03 F.D.Castel ***@***.***>:
… *fdcastel* left a comment (FirebirdSQL/firebird-odbc-driver#275)
<#275 (comment)>
#207 <#207> as
mentioned in this comment
<#275 (comment)>
Thanks, Mark! My oversight.
P.S. This tool <https://github.com/edwig/ODBCQueryTool> would be
enormously useful!
—
Reply to this email directly, view it on GitHub
<#275 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAKZQLWKQNGWR4IX7MOKY2D4LSP5VAVCNFSM6AAAAACUVFLU3CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTQOJRHAYDANZWGQ>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
|
I think that this might be a little too much change for a single pull request (but ultimately, that is for @irodushka to judge) |
Since this change was made early in the project -- before we removed dead code and performed other optimizations -- unfortunately, that’s correct. If you strongly prefer keeping the original implementation, we can certainly revisit that approach. It would require additional time, but it’s doable. That said, I believe the most effective path forward is to build a new parser and, with the author's permission, leverage the test suite from ODBCQueryTool/SQLComponents. |
It SURELY is. Honestly, I’m not entirely happy with how things turned out either. As I mentioned in the original post, I initially set out to fix just a few small pieces. However, it quickly evolved into something much larger, to the point where splitting it into smaller PRs was no longer feasible. My two cents: I’m not attached to the specific implementation. What matters to me is passing tests and maintaining solid code coverage. As long as those are in place, the code itself can be refactored or replaced as needed. |
|
Believe me - I really appreciate your job and I've called it "great" wo any
irony (well, maybe just a bit;) - 1% of irony vs 99% of respect).
From the other hand - I have to take care about the driver stability, and,
of course, the more massive changes are, the more dangerous and risky they
look like.
I think we should elaborate some compromise approach and find out a way to
minimize the risks, moving from revolutionary changes to evolutionary ones.
I'm sure we can handle it.
But not today, it's a dead of night here:)
Regards
чт, 12 февр. 2026 г., 22:30 F.D.Castel ***@***.***>:
… *fdcastel* left a comment (FirebirdSQL/firebird-odbc-driver#275)
<#275 (comment)>
I think that this might be a little too much change for a single pull
request (but ultimately, that is for @irodushka
<https://github.com/irodushka> to judge)
It SURELY is.
Honestly, I’m not entirely happy with how things turned out either.
As I mentioned in the original post, I initially set out to fix just a few
small pieces. However, it quickly evolved into something much larger, to
the point where splitting it into smaller PRs was no longer feasible.
My two cents: I’m not attached to the specific implementation — what
matters to me is passing tests and maintaining solid code coverage. As long
as those are in place, the code itself can be refactored or replaced as
needed.
—
Reply to this email directly, view it on GitHub
<#275 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAKZQLXU6T3KLN2FR4XRTH34LTIGPAVCNFSM6AAAAACUVFLU3CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTQOJSHE3DKMZUGE>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
|
I FULLY understand you, @irodushka -- I’d probably do the same if I were in your shoes. Enjoy the festivities! I’ll brainstorm something on my end during Carnival, here! 😄 |
|
I've been thinking about how we should move forward with this PR, and I'd like to propose the following approach. As always, opinions, criticism, and suggestions are very welcome. General Strategy
Initial BreakdownThe following outlines an initial plan for how we should approach this problem. Each item will be submitted as a separate PR, potentially structured as a sequence of dependent commits (building on one another). 1. Build System
2. Test Suite
3. GitHub Actions automation
The changes above are the most necessary. Without them, the remaining tasks would require significant manual testing and verification effort, making progress inefficient and resource-intensive. 4. Next Steps: Following the Original Plan -- Without the Bad Parts (*)From this point on, we largely follow the original roadmap, excluding the problematic decisions:
5. Final Cleanup (Very Last Step)
|
|
Hi @fdcastel - agree with you plan. This is exactly what I meant yesterday between the lines of my message%) What do you think about the following plan? - DevOps phaseSo, in the first step we should get a test suite (with some test cases that are expected to fail) and it should be run manually. The 2nd step - a new build system. Well, I dont think removing the last actual sln (MsVc2022.win folder) is reasonable. All other *.win solutions - ok, they are outdated, but MsVc2022.win should be kept for the customers who prefer to build from the native Visual Studio sln. And the 3rd step - a new CI (GitHib actions). At this stage, we move from manual launch of test suite to automated. Bugfix phaseHere we should fix the known errors and misbehavior cases. Every case must be issued as a separate ticket, branch and PR. Improvement phase (new features, refactoring & optimizations)[to do closer to the point] |
OdbcJdbcSetup stuff is for Windows only. Without that dialog you have a chance to setup ODBC by RegEdit, but I never give you such an advice)) On Linux there is ODBCConfig tool (Qt-based, as you like, @fdcastel %) but I never use it, preferring to configure the ODBC drivers/DSNs manually in /etc/ Frankly speaking, have no idea how it works on Mac. I'd guess the unixODBC approach would work, given macOS's Unix base. |
3 participants
v4 - Firebird ODBC Driver Modernization
Releasessection of the GitHub project. #231Builds available at https://github.com/fdcastel/firebird-odbc-driver/releases/tag/v4.0.1-preview.
Summary
This PR is a comprehensive modernization of the Firebird ODBC driver, addressing crash bugs, ODBC specification violations, Unicode data corruption on Linux/macOS, and significant performance bottlenecks. It also replaces the build system with CMake, adds a 401-test Google Test suite, and reorganizes the source tree.
The changes are organized into 4 commits for easier review:
build: remove legacy build files, vendored headers, and dead code-- 383 files removed. Safe to skim; this is just deletions.feat: modernize driver -- CMake build, source reorg, fixes and improvements-- The core changes. Source moved tosrc/, all driver improvements applied. Git detects 137 renames sogit diff -Mshows only the actual code changes (~5,300 lines of real modifications).test: add comprehensive Google Test suite (401 tests, 34 suites)-- New tests only. Can be reviewed independently.docs: add project documentation and developer guidelines-- Documentation only.Motivation
I started this project to address real bugs I was hitting while using the Firebird ODBC driver in production. What began as a few targeted fixes grew into a larger effort as I kept finding interconnected issues -- the crash bugs led me to the GUARD macro pattern, which led me to entry-point hardening, which revealed the missing SQLSTATE mappings, and so on.
The driver has a solid foundation -- it correctly uses the Firebird OO API for connections, transactions, statements, and result sets. But it has accumulated technical debt over its 20+ year history, and some areas (Unicode handling, error mapping, safety at the API boundary) needed significant work.
I understand this is a large PR and may take time to review. I'm happy to discuss any changes, split things differently, or revert specific decisions if they don't align with the project's direction.
This work also led to a secondary project: ODBC Crusher. What began as a tool to investigate and learn more about ODBC drivers has since grown into a fully usable tool (still under active development).
What Changed (and Why)
🔴 Crash Fixes
GUARD_HDESC/GUARD_HSTMT/GUARD_HDBCmacros dereference handles before checking for nullNULL_CHECKmacro -- check first, then dereferenceSQLCopyDesccrashes when source descriptor has no recordssour.records+ early return whenheadCount == 0OdbcObject::postErrorusessprintfinto 256-byte stack buffersnprintfwith 512-byte buffer(SQLException&)excatch (SQLException& exception)clausesSQL_INVALID_HANDLE🟡 ODBC Specification Compliance
SQL_NO_DATA_FOUND(ODBC 2.x constant) instead ofSQL_NO_DATA.StringLengthPtrwithout null check.NULL InfoValuePtr.SQL_SUCCESSwhen no cursor was open (should be SQLSTATE 24000).SQL_CHAR/SQL_VARCHARinstead ofSQL_WCHAR/SQL_WVARCHAR.SQLGetEnvAttralways returnedSQL_OV_ODBC3regardless of what was set.supportsCatalogsInIndexDefinitions()instead of the schema equivalent.SQL_OV_ODBC3_80now accepted; driver reports"03.80";SQL_ATTR_RESET_CONNECTION,SQL_GD_OUTPUT_PARAMS,SQL_ASYNC_DBC_FUNCTIONSimplemented.SQL_AM_STATEMENTwhile rejecting async enable with HYC00. Now correctly reportsSQL_AM_NONE.🟢 New Features
autoCommit=OFF. Failed statements don't corrupt the transaction.IBatchfor a single server roundtrip instead of N individual roundtrips. Falls back to row-by-row for Firebird 3.x. Inline BLOB support viaBLOB_ID_ENGINEpolicy.CHAR(16) CHARACTER SET OCTETS(FB3) andBINARY(16)(FB4+) are mapped toSQL_GUID. All conversion methods implemented.SQL_ATTR_QUERY_TIMEOUTimplemented usingIAttachment::cancelOperation(fb_cancel_raise)with a timer thread. Returns SQLSTATEHYT00on expiry.cancelOperation()to interrupt in-flight operations.getServerMajorVersion()/getServerMinorVersion()for version-gated behavior (e.g., FB4+ types like INT128, DECFLOAT, BINARY/VARBINARY).SQL_ATTR_RESET_CONNECTIONrolls back pending transactions, closes open cursors, and resets statement attributes.⚡ Performance
Mutex(CreateMutex/WaitForSingleObject) with user-modeSRWLOCKon Windows. ~20ns vs ~1-2μs per lock/unlock.IscResultSetfetches 64 rows at a time from Firebird, serving subsequentSQLFetchcalls from the buffer. Measured: 10.88 ns/row for 10K×10 INT columns (embedded).OdbcStringw-cache inDescRecord--SQLDescribeColW,SQLColAttributeW,SQLGetDescFieldW,SQLGetDescRecW,SQLGetDiagRecW,SQLGetDiagFieldWall read from cached UTF-16 strings viamemcpy, bypassingConvertingStringentirely.Value::getStringbuffer reuse,clearErrorsfast path,std::to_charsfor float→string,ODBC_FORCEINLINEon hot helpers,alignas(64)onSqlda::buffer.🔤 Unicode
The original Unicode implementation assumed
SQLWCHAR == wchar_t, which is only true on Windows. On Linux (wchar_t= 4 bytes), this caused complete data corruption for all W-API functions.ODBC_SQLWCHARtypedef (always 16-bit on all platforms).*ToStringWconversion functions now useSQLWCHAR*directly.Utf16Convert.h).CHARSETdefaults toUTF8when not specified in the connection string (wasNONE, which used locale-dependentmbstowcson Linux).convVarStringSystemToStringWfixed to useSQLWCHAR*instead ofwchar_t*.🏗️ Build System & Infrastructure
cmake/GetVersionFromGit.cmake.installer/Product.wxsfor Windows deployment.src/andsrc/IscDbc/.🧹 Code Quality
OdbcErrorchain:std::vector<std::unique_ptr<OdbcError>>replaces raw linked list.IscConnection::statements,IscStatement::resultSets,IscBlob::blobs:std::vectorreplacesLinkedList.OdbcObject,OdbcError,OdbcEnv: diagnostic fields madeprivate/protected.EnvShare: Meyer's Singleton (construct-on-first-use) fixes static initialization order.IscDbcclasses markedfinalfor devirtualization.🗑️ What Was Removed
Builds/(14 configs)FBClient.Headers/(60 files)Headers/(SQL.H, SQLEXT.H)OdbcJdbcSetup/(30 files)Install/(HTML help, InnoSetup)Res/(5 locale files)Test/,Tests/,JdbcTest/,TestInstall/ChangeLogfilesTesting
GTEST_SKIP()) when no database connection is available.Breaking Changes
src/andsrc/IscDbc/.CHARSETnow defaults toUTF8(wasNONE). This is a correctness improvement --CHARSET=NONEwith locale-dependentmbstowcswas producing incorrect results on non-UTF-8 locales.{fn},{d},{ts},{oj}escape processing has been removed. SQL is sent as-is to Firebird. Applications using ODBC escapes should switch to native Firebird SQL syntax.odbcconforodbcinstfor driver registration.Transparency Note
This work was substantially assisted by AI (LLMs). I believe in being upfront about this: the AI was used for code generation, analysis, testing, and documentation throughout this project. Every change was reviewed, tested against a real Firebird database, and verified on CI. The detailed issue tracking in Docs/FIREBIRD_ODBC_MASTER_PLAN.md documents the reasoning behind each change.
I believe the results speak for themselves: 401 passing tests, crash bugs eliminated, correct SQLSTATE mapping, functional Unicode on Linux, and measurable performance improvements. But I understand if the use of AI tooling is a concern, and I'm happy to discuss any specific changes in detail.