TRY_CONVERT by robozmey · Pull Request #901 · apache/cloudberry

robozmey · 2025-02-03T13:18:54Z

Extension TRY_CONVERT adds function TRY_CONVERT

TRY_CONVERT was made to use like TRY_CAST from SQL-Server

Due polymorphism have strange signature (some_type, anyelement) -> anyelement

First parament is SOURCE_VALUE

Second parameter is DEFAULT_VALUE::TARGET_TYPE

TRY_CONVERT('42'::text, NULL::int2) -- returns 42::int2
TRY_CONVERT('42d'::text, NULL::int2) -- returns NULL::int2
TRY_CONVERT('42d'::text, 1234::int2) -- returns 1234::int2

We cannot have (any1, any2) -> any2 signature with different anyelement types, but we can create multiple (_, any) -> any functions in contrib/try_convert/try_convert--1.0.sql or use select add_type_for_try_convert('date'::regtype); to add types.

In init we can add all supported types, but with add_type_for_try_convert user can add whatever type he wants

ERROR HANDLING

Error handles by PG_TRY() PG_CATCH()

But better way is use "soft" error handling from Postgres 17 (postgres/postgres@ccff2d2) it also realising OPENGPDB (open-gpdb/gpdb@21be368). Of course it needs to convert datatype function to support "soft" error handling

Fixes #ISSUE_Number

What does this PR do?

Type of Change

Bug fix (non-breaking change)
New feature (non-breaking change)
Breaking change (fix or feature with breaking changes)
Documentation update

Breaking Changes

Test Plan

Unit tests added/updated
Integration tests added/updated
Passed make installcheck
Passed make -C src/test installcheck-cbdb-parallel

Impact

Performance:

User-facing changes:

Dependencies:

Checklist

Followed contribution guide
Added/updated documentation
Reviewed code for security implications
Requested review from cloudberry committers

Additional Context

CI Skip Instructions

gfphoenix78 · 2025-06-03T10:07:24Z

+	if (OidIsValid(sourceTypeId))
+		sourceTypeId = getBaseType(sourceTypeId);
+	if (OidIsValid(targetTypeId))
+		targetTypeId = getBaseType(targetTypeId);


What if the source or target type is invalid?

All will happend like in

cloudberry/src/backend/parser/parse_coerce.c

Line 3151 in e601fb6

if (OidIsValid(sourceTypeId))

(My conversion algorithm duplicates internal one)

We will look in pg_cast table - will find nothing, will try to get TypeCategory and will get assert in TypeCategory(targetTypeId)

gfphoenix78 · 2025-06-03T10:21:56Z

+
+		*funcId = castForm->castfunc;
+		ReleaseSysCache(tuple);
+	}


Better add else branch here.

gfphoenix78 · 2025-06-03T10:24:21Z

+}
+
+Datum
+convert_from_function(Datum value, int32 typmod, Oid funcId, bool *is_failed)


Why is the error handling undetermined?

* Add README for try_convert extension

tuhaihe · 2025-07-17T08:22:42Z

@@ -0,0 +1,92 @@
+# TRY_CONVERT
+
+TRY_CONVERT is Greenplum extension, which adds function for error-safe type cast like [TRY_CAST from SQL-Server](https://learn.microsoft.com/ru-ru/sql/t-sql/functions/try-cast-transact-sql?view=sql-server-ver16)


Suggested change

TRY_CONVERT is Greenplum extension, which adds function for error-safe type cast like [TRY_CAST from SQL-Server](https://learn.microsoft.com/ru-ru/sql/t-sql/functions/try-cast-transact-sql?view=sql-server-ver16)

TRY_CONVERT is Greenplum/Cloudberry extension, which adds function for error-safe type cast like [TRY_CAST from SQL-Server](https://learn.microsoft.com/ru-ru/sql/t-sql/functions/try-cast-transact-sql?view=sql-server-ver16)

tuhaihe · 2025-07-17T08:23:01Z

+
+## Why signature is so strange?
+
+Greenplum function polymorphism accept to have polymorphic functions only one any type in signature.  


Suggested change

Greenplum function polymorphism accept to have polymorphic functions only one any type in signature.

Greenplum/Cloudberry function polymorphism accept to have polymorphic functions only one any type in signature.

tuhaihe · 2025-07-17T08:26:15Z

Hi @robozmey thanks for your updates! Now the license headers look good to me. Please make sure these files are originally created by our community members, not cherry-picked or copied from other projects.

Before merging, welcome to squash your commits into one and revise the commit message body, you can take this template as a reference.

TRY_CONVERT(source_value, default_value) casts source_value to the type of default_value and returns default_value whenever the cast fails, which is the behaviour of TRY_CAST in SQL Server. Without it a single malformed value makes the whole query fail, and the plpgsql workarounds that catch the error per row are several times slower. TRY_CONVERT('42'::text, NULL::int2) -- returns 42::int2 TRY_CONVERT('42d'::text, NULL::int2) -- returns NULL::int2 TRY_CONVERT('42d'::text, 1234::int2) -- returns 1234::int2 The conversion to use is resolved the same way the parser resolves an explicit cast in coerce_type(): a pg_cast entry, an I/O conversion or a binary-compatible relabel, followed by the length coercion of the target type. Casts between array types and casts to domain types are not supported and are reported as query errors, the same as a cast that does not exist at all; only a failure caused by the converted data is turned into the default value. The cast itself runs inside a PG_TRY() block, since the datatype input functions of Cloudberry cannot yet report a conversion failure without throwing. Query cancellation and assertion failures are re-thrown rather than swallowed, following what plpgsql does for "EXCEPTION WHEN others". Once the input functions are converted to the soft error handling infrastructure of PostgreSQL 17, the PG_TRY() block can go away. Polymorphism only accepts a single "any" type per signature, hence the default value carrying the target type. The built-in types are registered by the extension script; add_type_for_try_convert(regtype) registers additional ones, for example the hstore and citext types. The extension is marked trusted, so the owner of a database can install it without being a superuser. The regression test is derived from the catalog files and from the sample values in contrib/try_convert/data/, so "make installcheck" generates it instead of the tree carrying it. The ic-contrib job of the build workflows runs it. The extension was written by Vladimir Rachkin and proposed in PR#901; this commit rebases that work and addresses the review comments left on it. Co-authored-by: Vladimir Rachkin <vova@kpnn.ru> See: PR#901 <apache#901>

TRY_CONVERT(source_value, default_value) casts source_value to the type of default_value and returns default_value whenever the cast fails, which is the behaviour of TRY_CAST in SQL Server. Without it a single malformed value makes the whole query fail, and the plpgsql workarounds that catch the error per row are several times slower. TRY_CONVERT('42'::text, NULL::int2) -- returns 42::int2 TRY_CONVERT('42d'::text, NULL::int2) -- returns NULL::int2 TRY_CONVERT('42d'::text, 1234::int2) -- returns 1234::int2 The conversion to use is resolved the same way the parser resolves an explicit cast in coerce_type(): a pg_cast entry, an I/O conversion or a binary-compatible relabel, followed by the length coercion of the target type. Casts between array types and casts to domain types are not supported and are reported as query errors, the same as a cast that does not exist at all; only a failure caused by the converted data is turned into the default value. The cast itself runs inside a PG_TRY() block, since the datatype input functions of Cloudberry cannot yet report a conversion failure without throwing. The extension is marked trusted, so the owner of a database can install it without being a superuser. Co-authored-by: Vladimir Rachkin <vova@kpnn.ru> See: PR#901 <apache#901>

Alena0704 · 2026-07-30T11:58:01Z

I don't have push access to the original fork, and the head branch of an existing PR can't be changed, so instead of updating this one I opened a separate PR #1872 with the rebased version.

All credit for the extension goes to @robozmey - the commit lists him as a co-author.

On top of the rebase I addressed the open review comments, pulled in the follow-up fixes the extension got in open-gpdb since this PR was opened, added the module to the ic-contrib CI matrix (nothing was running its tests before), and marked the extension trusted so that a non-superuser with CREATE privilege on the database can install it.

TRY_CONVERT(source_value, default_value) casts source_value to the type of default_value and returns default_value whenever the cast fails, which is the behaviour of TRY_CAST in SQL Server. Without it a single malformed value makes the whole query fail, and the plpgsql workarounds that catch the error per row are several times slower. TRY_CONVERT('42'::text, NULL::int2) -- returns 42::int2 TRY_CONVERT('42d'::text, NULL::int2) -- returns NULL::int2 TRY_CONVERT('42d'::text, 1234::int2) -- returns 1234::int2 The conversion to use is resolved the same way the parser resolves an explicit cast in coerce_type(): a pg_cast entry, an I/O conversion or a binary-compatible relabel, followed by the length coercion of the target type. Casts between array types and casts to domain types are not supported and are reported as query errors, the same as a cast that does not exist at all; only a failure caused by the converted data is turned into the default value. The cast itself runs inside a PG_TRY() block, since the datatype input functions of Cloudberry cannot yet report a conversion failure without throwing. The extension is marked trusted, so the owner of a database can install it without being a superuser. Co-authored-by: Vladimir Rachkin <vova@kpnn.ru> See: PR#901 <apache#901>

tuhaihe · 2026-07-31T02:42:51Z

Maybe we can close this PR since we have a new one.

TRY_CONVERT(source_value, default_value) casts source_value to the type of default_value and returns default_value whenever the cast fails, which is the behaviour of TRY_CAST in SQL Server. Without it a single malformed value makes the whole query fail, and the plpgsql workarounds that catch the error per row are several times slower. TRY_CONVERT('42'::text, NULL::int2) -- returns 42::int2 TRY_CONVERT('42d'::text, NULL::int2) -- returns NULL::int2 TRY_CONVERT('42d'::text, 1234::int2) -- returns 1234::int2 The conversion to use is resolved the same way the parser resolves an explicit cast in coerce_type(): a pg_cast entry, an I/O conversion or a binary-compatible relabel, followed by the length coercion of the target type. Casts between array types and casts to domain types are not supported and are reported as query errors, the same as a cast that does not exist at all; only a failure caused by the converted data is turned into the default value. The cast itself runs inside a PG_TRY() block, since the datatype input functions of Cloudberry cannot yet report a conversion failure without throwing. The extension is marked trusted, so the owner of a database can install it without being a superuser. Co-authored-by: Vladimir Rachkin <vova@kpnn.ru> See: PR#901 <apache#901>

TRY_CONVERT(source_value, default_value) casts source_value to the type of default_value and returns default_value whenever the cast fails, which is the behaviour of TRY_CAST in SQL Server. Without it a single malformed value makes the whole query fail, and the plpgsql workarounds that catch the error per row are several times slower. TRY_CONVERT('42'::text, NULL::int2) -- returns 42::int2 TRY_CONVERT('42d'::text, NULL::int2) -- returns NULL::int2 TRY_CONVERT('42d'::text, 1234::int2) -- returns 1234::int2 The conversion to use is resolved the same way the parser resolves an explicit cast in coerce_type(): a pg_cast entry, an I/O conversion or a binary-compatible relabel, followed by the length coercion of the target type. Casts between array types and casts to domain types are not supported and are reported as query errors, the same as a cast that does not exist at all; only a failure caused by the converted data is turned into the default value. The cast itself runs inside a PG_TRY() block, since the datatype input functions of Cloudberry cannot yet report a conversion failure without throwing. The extension is marked trusted, so the owner of a database can install it without being a superuser. Co-authored-by: Vladimir Rachkin <vova@kpnn.ru> See: PR#901 <#901>

gfphoenix78 reviewed Jun 3, 2025

View reviewed changes

Comment thread contrib/try_convert/try_convert.c

gfphoenix78 reviewed Jun 3, 2025

View reviewed changes

Comment thread contrib/try_convert/try_convert.c Outdated

gfphoenix78 reviewed Jun 3, 2025

View reviewed changes

Comment thread contrib/try_convert/try_convert.c Outdated

gfphoenix78 reviewed Jun 3, 2025

View reviewed changes

robozmey force-pushed the try_convert-in-cb branch from 7c1d1e5 to c3b4c2b Compare June 11, 2025 09:21

tuhaihe reviewed Jul 15, 2025

View reviewed changes

Comment thread contrib/try_convert/Makefile

Comment thread contrib/try_convert/scripts/check_test.py

Comment thread contrib/try_convert/try_convert--1.0.sql

Comment thread contrib/try_convert/try_convert.c

robozmey added 6 commits July 16, 2025 11:57

Add TRY_CONVERT extension

9a06b75

Remove unsuported types

cb7d0c0

Fix tab indents

4689cbe

Make internal functions static

9a4bc17

Refactor find_typmod_conversion_function

c4c7afa

Add Licensies

fde465f

robozmey force-pushed the try_convert-in-cb branch from 737228d to fde465f Compare July 16, 2025 08:57

robozmey added 2 commits July 16, 2025 12:01

Add licence for .c file

454bd4c

Add README for try_convert extension (apache#179)

df9a4d5

* Add README for try_convert extension

tuhaihe reviewed Jul 17, 2025

View reviewed changes

Alena0704 mentioned this pull request Jul 30, 2026

Feature: add TRY_CONVERT extension #1872

Merged

13 tasks

This was referenced Jul 30, 2026

Feature: add TRY_CONVERT extension #1873

Open

Feature: add TRY_CONVERT extension open-gpdb/cloudberry#51

Open

tuhaihe closed this Jul 31, 2026

		@@ -0,0 +1,92 @@
		# TRY_CONVERT

		TRY_CONVERT is Greenplum extension, which adds function for error-safe type cast like [TRY_CAST from SQL-Server](https://learn.microsoft.com/ru-ru/sql/t-sql/functions/try-cast-transact-sql?view=sql-server-ver16)

	TRY_CONVERT is Greenplum extension, which adds function for error-safe type cast like [TRY_CAST from SQL-Server](https://learn.microsoft.com/ru-ru/sql/t-sql/functions/try-cast-transact-sql?view=sql-server-ver16)
	TRY_CONVERT is Greenplum/Cloudberry extension, which adds function for error-safe type cast like [TRY_CAST from SQL-Server](https://learn.microsoft.com/ru-ru/sql/t-sql/functions/try-cast-transact-sql?view=sql-server-ver16)


		## Why signature is so strange?

		Greenplum function polymorphism accept to have polymorphic functions only one any type in signature.

	Greenplum function polymorphism accept to have polymorphic functions only one any type in signature.
	Greenplum/Cloudberry function polymorphism accept to have polymorphic functions only one any type in signature.

Uh oh!

Conversation

robozmey commented Feb 3, 2025

Extension TRY_CONVERT adds function TRY_CONVERT

ERROR HANDLING

What does this PR do?

Type of Change

Breaking Changes

Test Plan

Impact

Checklist

Additional Context

CI Skip Instructions

Uh oh!

Uh oh!

Uh oh!

Uh oh!

gfphoenix78 Jun 3, 2025

Choose a reason for hiding this comment

Uh oh!

robozmey Jun 11, 2025

Choose a reason for hiding this comment

Uh oh!

gfphoenix78 Jun 3, 2025

Choose a reason for hiding this comment

Uh oh!

gfphoenix78 Jun 3, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

tuhaihe Jul 17, 2025

Choose a reason for hiding this comment

Uh oh!

tuhaihe Jul 17, 2025

Choose a reason for hiding this comment

Uh oh!

tuhaihe commented Jul 17, 2025

Uh oh!

Alena0704 commented Jul 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tuhaihe commented Jul 31, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Alena0704 commented Jul 30, 2026 •

edited

Loading