Reviewing files that changed from the base of the PR and between e9d96a8 and 289f0cc.

• .github/workflows/build.yml (2 hunks)
• contrib/gb18030_2022/utf8_and_gb18030_2022.c (1 hunks)
• contrib/ivorysql_ora/src/builtin_functions/datetime_datatype_functions.c (3 hunks)
• contrib/uuid-ossp/uuid-ossp.c (2 hunks)
• src/backend/access/transam/xlog.c (1 hunks)
• src/backend/catalog/genbki.pl (1 hunks)
• src/backend/commands/sequence.c (2 hunks)
• src/backend/commands/tablecmds.c (2 hunks)
• src/backend/executor/execExpr.c (1 hunks)
• src/backend/oracle_parser/liboracle_parser.c (1 hunks)
• src/backend/parser/parse_param.c (1 hunks)
• src/backend/utils/adt/arrayfuncs.c (1 hunks)
• src/backend/utils/adt/ruleutils.c (2 hunks)
• src/backend/utils/misc/ivy_guc.c (1 hunks)
• src/bin/pg_dump/pg_dump.c (1 hunks)
• src/bin/psql/common.c (5 hunks)
• src/bin/psql/psqlplusscan.l (1 hunks)
• src/bin/psql/startup.c (1 hunks)
• src/fe_utils/string_utils.c (1 hunks)
• src/include/utils/ora_compatible.h (1 hunks)
• src/interfaces/libpq/ivy-exec.c (9 hunks)
• src/oracle_test/isolation/isolation_main.c (1 hunks)
• src/oracle_test/isolation/specparse.y (1 hunks)
• src/pl/plisql/src/pl_comp.c (1 hunks)
• src/pl/plisql/src/pl_handler.c (1 hunks)
• src/pl/plisql/src/pl_package.c (1 hunks)
• src/pl/plisql/src/pl_scanner.c (2 hunks)
• src/pl/plisql/src/pl_subproc_function.c (1 hunks)

• GitHub Check: pg_regression (ubuntu-latest)
• GitHub Check: build (ubuntu-latest)
• GitHub Check: meson_build (ubuntu-latest)
• GitHub Check: contrib_regression
• GitHub Check: oracle_pg_regression (ubuntu-latest)
• GitHub Check: oracle_regression (ubuntu-latest)

src/fe_utils/string_utils.c (1)

107-115: LGTM! Resource allocation optimized.

The reordering defers buffer allocation until after the Oracle mode check, avoiding unnecessary work when the early return is taken. This change eliminates wasted resource allocation and likely resolves a GCC 14.2 compiler warning about potentially unused initialization.

src/backend/utils/adt/arrayfuncs.c (1)

468-468: LGTM: Minor formatting improvement.

This whitespace adjustment aligns with the PR's objective to address build warnings with GCC 14.2. The error handling logic remains correct and consistent with the similar error path at line 454.

contrib/uuid-ossp/uuid-ossp.c (1)

565-568: LGTM! Clean refactoring to eliminate build warnings.

Consolidating the OSSP-specific variable declarations at the top of the function with proper conditional compilation guards addresses both "declaration after statement" and "unused variable" warnings. This change aligns with the PR's goal of enabling stricter GCC 14 compilation flags.

src/oracle_test/isolation/specparse.y (1)

18-19: Verify header file placement for extern declarations.

These extern declarations for Bison-generated parser internals (spec_yychar and spec_yynerrs) are placed in the .y file prologue. If these variables are accessed from other compilation units, they should be declared in a header file (e.g., specparse.h) rather than in the grammar file prologue, as the latter only affects the generated parser source file. Verify whether these variables are used externally and consider moving their declarations to an appropriate header file for consistency with C best practices.

src/backend/executor/execExpr.c (1)

5108-5120: Initialize func_tuple to satisfy GCC 14 without changing behavior

Initializing HeapTuple func_tuple = NULL; is safe and removes the GCC 14 -Wmaybe-uninitialized warning. func_tuple is only assigned and released inside the FUNC_EXPR_FROM_PG_PROC() branch, so this has no semantic impact and just quiets the new stricter warnings.

.github/workflows/build.yml (1)

21-41: GCC 14 setup and stricter CFLAGS are consistent with the PR goals

Installing gcc-14/g++-14 and wiring them via update-alternatives on ubuntu-latest (Ubuntu 24.04, which provides gcc-14 14.2.0-4ubuntu2~24.04) is valid, and the added CFLAGS warning/error options align with tightening CI under GCC 14. The sequence of installing gcc-14 first, then build-essential, with higher alternative priority for gcc-14, should keep 14.x as the active compiler.

If you later extend the os matrix beyond Ubuntu, ensure this step remains guarded by the ubuntu-latest check, and be prepared for some existing code paths to start failing CI due to the new -Werror=* flags.

src/backend/commands/sequence.c (1)

98-101: Restrict session_id and scale_value to file scope

Making session_id (int64) and scale_value (int32) static confines them to this translation unit while preserving zero-initialization. This aligns with -Wmissing-variable-declarations and visibility best practices. Behavior is unchanged if these variables are only used within sequence.c.

Before merging, confirm that no other translation unit declares or references session_id or scale_value.

src/pl/plisql/src/pl_comp.c (1)

1097-1116: Arg/return type reference tracking looks correct and self‑contained

Resetting argtypenames/rettypename before calling get_func_typename_info and then walking the arrays to feed plisql_add_type_referenced_objects is logically sound and matches how %TYPE/%ROWTYPE references are handled elsewhere. The strcmp(argtypenames[j], "") != 0 guard also ensures you skip empty slots safely.

No functional issues here from a correctness or lifetime perspective (given the subsequent pfree(argtypenames)/pfree(rettypename)).

src/pl/plisql/src/pl_handler.c (1)

228-234: _PG_fini prototype is appropriate

Adding an explicit void _PG_fini(void); prototype before the definition is correct for a non‑static module lifecycle function and will satisfy stricter compiler checks without changing behavior.

src/bin/psql/psqlplusscan.l (1)

50-50: Static NumKeywords is fine and reduces symbol export

Restricting NumKeywords to internal linkage with static matches its usage (only referenced in this file) and avoids exporting an unnecessary global.

src/include/utils/ora_compatible.h (1)

61-61: New extern for bootstrap_database_mode is reasonable; ensure single definition

Declaring extern int bootstrap_database_mode; here is fine given this header’s role in compatibility mode plumbing. Please just verify there is exactly one non‑static definition of bootstrap_database_mode in the backend (and no leftover conflicting externs in other headers) to avoid link‑time issues.

src/backend/oracle_parser/liboracle_parser.c (1)

52-52: Static prev_quote_identifier matches other saved‑hook variables

Marking prev_quote_identifier as static is correct; it’s only used inside this module to save/restore the hook and no longer pollutes the global symbol table.

src/backend/parser/parse_param.c (1)

74-75: Making TopOraParamNode/CurrentOraParamNode static is appropriate

These stacks are only manipulated via the public helper functions in this file, so changing them to static is a good encapsulation step and should silence global‑symbol warnings. However, verify that no other .c file still references these symbols directly by searching the codebase.

src/bin/pg_dump/pg_dump.c (1)

19115-19131: SEQTYPE_CUSTINT is never handled and now always triggers pg_fatal

SeqType defines SEQTYPE_CUSTINT and SeqTypeNames[] maps it to "pg_catalog.int8", but the default range computation in dumpSequence only handles SEQTYPE_SMALLINT, SEQTYPE_INTEGER, and SEQTYPE_BIGINT. Any sequence whose type parses as SEQTYPE_CUSTINT will now fall into the final else and abort pg_dump with "unrecognized sequence type", even though it is explicitly recognized by parse_sequence_type.

If SEQTYPE_CUSTINT is intended to be an alias of bigint (which the "pg_catalog.int8" mapping strongly suggests), it should be treated like SEQTYPE_BIGINT here rather than failing.

Consider updating the conditional to group SEQTYPE_CUSTINT with the bigint case and keep the pg_fatal only for truly unknown enum values:

-    else if (seq->seqtype == SEQTYPE_BIGINT)
-    {
-        default_minv = is_ascending ? 1 : PG_INT64_MIN;
-        default_maxv = is_ascending ? PG_INT64_MAX : -1;
-    }
-    else
-    {
-        pg_fatal("unrecognized sequence type: %d", seq->seqtype);
-        default_minv = default_maxv = 0;    /* keep compiler quiet */
-    }
+    else if (seq->seqtype == SEQTYPE_BIGINT ||
+             seq->seqtype == SEQTYPE_CUSTINT)
+    {
+        default_minv = is_ascending ? 1 : PG_INT64_MIN;
+        default_maxv = is_ascending ? PG_INT64_MAX : -1;
+    }
+    else
+    {
+        pg_fatal("unrecognized sequence type: %d", seq->seqtype);
+        default_minv = default_maxv = 0;    /* keep compiler quiet */
+    }

Please verify that sequences with seqtypid formatting as "pg_catalog.int8" are handled correctly.

src/backend/access/transam/xlog.c (1)

112-114: BootstrapTimeLineID macro looks correct and improves clarity

Defining BootstrapTimeLineID as a single macro (set to 1) matches its usage in the bootstrap WAL/control‑file paths and removes the hard‑coded magic value without changing behavior. No further changes needed here.

contrib/gb18030_2022/utf8_and_gb18030_2022.c (1)

25-26: Static converter hooks look correct

Making pre_gb18030_2022_to_utf8_hook and pre_utf8_to_gb18030_2022_hook static correctly restricts them to this translation unit without changing behavior, since they’re only used in _PG_init/_PG_fini.

src/oracle_test/isolation/isolation_main.c (1)

21-23: Static globals for isolation tester path are fine

saved_argv0, isolation_exec, and looked_up_isolation_exec are only used in this file, so making them static is a safe visibility tightening and should not affect behavior.

src/bin/psql/common.c (1)

3726-3735: Initializing i in SendQuery_PBE is harmless and silences warnings

Changing int i; to int i = 0; in SendQuery_PBE is behavior‑preserving (all uses of i are in for (i = …) loops) and should quiet any “maybe uninitialized” warnings from GCC 14 around the sendquery_cleanup label.

src/bin/psql/startup.c (1)

399-440: Explicit NULL initialization for scan_state/cond_stack is good

Initializing PsqlScanState scan_state = NULL; and ConditionalStack cond_stack = NULL; in the ACT_SINGLE_SLASH path avoids “maybe uninitialized” warnings and guarantees a known value if db_mode were ever something other than DB_PG/DB_ORACLE (in addition to the existing Assert(scan_state != NULL)).

src/pl/plisql/src/pl_package.c (1)

2732-2739: Improved assertion around package–function reference removal

Capturing pre_len = list_length(psource->source.funclist); before list_delete_ptr() and then asserting pre_len == list_length(...) + 1 is a correct and clearer way to verify that exactly one entry was removed. This also makes it easier to detect cases where func wasn’t present in the list.

src/backend/utils/misc/ivy_guc.c (1)

37-39: Making NLS backing variables static is reasonable; verify no external extern declarations

Restricting nls_territory, nls_currency, and nls_iso_currency to internal linkage aligns with how they're used here (exclusively through the GUC machinery in this file) and reduces symbol exposure.

Ensure there are no extern declarations or direct external references to these variables elsewhere in the codebase—if other files reference them directly rather than through GetConfigOption*, linking will fail after this change.

src/backend/commands/tablecmds.c (1)

15670-15673: Good defensive initialization of dependentFuncPkgOids.

Initializing ObjectFunOrPkg *dependentFuncPkgOids to NULL removes the “maybe-uninitialized” risk and makes subsequent control-flow easier to reason about; no issues here.

contrib/ivorysql_ora/src/builtin_functions/datetime_datatype_functions.c (1)

156-204: Static linkage for lookup tables looks good

Making ora_days, date_fmt, and date_timezone static correctly limits their visibility to this translation unit and should help with GCC 14 warnings about unused globals, without changing behavior.

src/interfaces/libpq/ivy-exec.c (1)

2997-3044: IvyaddValueToList initialization/scan logic is correct

Initializing pre, tmp, and find and using the for (tmp = list, pre = list; ...) pattern preserves the original “dedupe then append at tail” behavior while eliminating potential uninitialized‑use warnings. No functional issues here.

Reviewing files that changed from the base of the PR and between bc5192b and a1a2431.

• src/bin/psql/common.c (3 hunks)
• src/interfaces/libpq/ivy-exec.c (10 hunks)

• src/interfaces/libpq/ivy-exec.c

• GitHub Check: build (ubuntu-latest)
• GitHub Check: oracle_pg_regression (ubuntu-latest)
• GitHub Check: meson_build (ubuntu-latest)
• GitHub Check: oracle_regression (ubuntu-latest)
• GitHub Check: contrib_regression
• GitHub Check: pg_regression (ubuntu-latest)

src/bin/psql/common.c (2)

3852-3863: LGTM! Inline loop declaration and backward iteration are correct.

The inline int i declaration follows modern C best practices and limits scope. The backward iteration correctly accesses indices from hv->length - 1 down to 0 using i - 1, and the logic aligns with the Oracle error message ordering requirement mentioned in the comment.

---

4040-4044: LGTM! Inline loop declaration improves code clarity.

The inline int i declaration properly scopes the loop variable and follows C99+ best practices. The loop bounds are correct.