Add IDbTypeResolver to allow plugins to control how DbTypes are mapped by PauloHMattos · Pull Request #6267 · npgsql/npgsql

PauloHMattos · 2025-10-25T02:26:45Z

roji · 2025-10-25T15:53:14Z

NinoFloris

Thanks for working on this so quickly, this is a good start.

Take a look at my feedback on the files themselves. Additionally I'd like to see a bit of test coverage of this feature for the DbType and DataTypeName properties.

The tests should assert that when these properties are expected to return resolved results they do so correctly. After ResolveTypeInfo has run a user should be able to observe that their DataTypeName for DbType.String changed from e.g. "text" to "citext".

src/Npgsql/Internal/IDbTypeResolver.cs

src/Npgsql/Internal/PgSerializerOptions.cs

src/Npgsql/Internal/PgTypeInfoResolverChainBuilder.cs

src/Npgsql/NpgsqlParameter.cs

src/Npgsql/Internal/PgSerializerOptions.cs

src/Npgsql/NpgsqlParameter.cs

test/Npgsql.Tests/TypeMapperTests.cs

src/Npgsql/NpgsqlParameter.cs

PauloHMattos · 2025-11-08T21:12:28Z

test/Npgsql.Tests/TypeMapperTests.cs

+        Assert.That(parameter.NpgsqlDbType, Is.EqualTo(NpgsqlDbType.Uuid));
+        Assert.That(parameter.DataTypeName, Is.EqualTo("uuid"));


@NinoFloris here I think we want this:

Suggested change

Assert.That(parameter.NpgsqlDbType, Is.EqualTo(NpgsqlDbType.Uuid));

Assert.That(parameter.DataTypeName, Is.EqualTo("uuid"));

Assert.That(parameter.NpgsqlDbType, Is.EqualTo(NpgsqlDbType.Unknow));

Assert.That(parameter.DataTypeName, Is.EqualTo(type));

But since when NpgsqlParameter.DbType is set, _npgsqlDbType is also set I couldn't get this result.

To achieve this DbType would have to take priority in the get of NpgsqlDbType and DataTypeName, or remove this behavior of when one is set the other is also set.

I haven't test this properly but I suspect both approaches would cause regressions.
I will take a closer look at this tomorrow, but if you have any ideias, or see something wrong, let me know.

Yes they have their own storage now so they can be decoupled.

When both values are set, like in this test:

npgsql/test/Npgsql.Tests/Types/NumericTests.cs

Line 118 in 82a849f

await AssertTypeWrite(5.5m, "5.5", "numeric", NpgsqlDbType.Numeric, DbType.VarNumeric, inferredDbType: DbType.Decimal);

We should prioritize the value of NpgsqlDbType? Or let DbType return the value that was explicitly set into it.

Looks like most of the failing tests are related with the inference of the DbType when NpgsqlDbType and DbType are provided

Those should be fixed now. If you have anything else to add please let me know. Otherwise it all looks good to me now and I can merge it for 10.0.

@NinoFloris thank you so much! All looks good to me too.

Thanks for your work on this!

Fyi, I did just find an issue with the api shape and how it interacts with type reloading, which I didn't quite foresee.

Compared to the lifetime of the rest of the resolver chain (which is generally built once), in an IDbTypeResolver you'd ideally want to be able to use facts about the current database. For example, to only map to citext if it's actually installed. Doing that works in the current design if you do this on every call into the resolver, but it would be pretty suboptimal to do so. Instead you would want to cache that fact inside the IDbTypeResolver instance. However at that point when a user calls ReloadTypes we have no way of getting that cached state reset, as we cannot recreate the resolver chain today (and neither would we want to for the type info resolvers).

So, at a minimum I'll have to change how we track these db type resolvers such that we can recreate when required. But it's likely I'll also tweak the api signatures a bit as we may not need to flow options through the method arguments any longer.

EDIT:
Additionally, lumping this resolver in with the others on PgTypeInfoResolverFactory means we should technically also support this feature through the global type mapper. This is not desired as we don't want to add new functionality to it any longer. All of this points to making DbTypeResolverFactory its own thing.

src/Npgsql/NpgsqlParameter.cs

PauloHMattos · 2025-11-12T16:55:59Z

@NinoFloris Just wanted to let you know that I’ve been testing my applications with Npgsql from this branch with Babelfish, and everything is working flawlessly.

I honestly didn’t expect this change to require so much refactoring on your side — thank you so much for taking the time to review and improve it. I really appreciate all your work on this project!

NinoFloris · 2025-11-14T14:57:34Z

Good to hear it all works as expected.

Features in this area always end up being more involved than anticipated. The postgres plugin system really does make npgsql quite a bit more complex when it comes to type mapping than other drivers. So thank you to for taking a shot at a complex issue, hopefully we'll see you back again!

NinoFloris · 2025-11-16T19:24:00Z

src/Npgsql/NpgsqlCommandBuilder.cs

        var param = (NpgsqlParameter)p;
-        param.NpgsqlDbType = (NpgsqlDbType)row[SchemaTableColumn.ProviderType];
+        // DbCommandBuilder is going to set DbType.Int32 onto an existing parameter, reset other db type fields.
+        if (param.SourceColumnNullMapping)


See https://github.com/dotnet/runtime/blob/8f1dc6d3d2ed2918f1c68e285a5f3970e0d4cc91/src/libraries/System.Data.Common/src/System/Data/Common/DBCommandBuilder.cs#L1121. This is the only place that sets this property in the BCL. Why did this have to require heap storage on all parameter instances and couldn't be passed through this ApplyParameterInfo method? Who knows, it's probably very necessary™️

src/Npgsql/NpgsqlDataSourceBuilder.cs

… DbTypes are mapped

This showed up in 'Timestamptz_as_DateTimeOffset_utc_with_DbType_DateTimeOffset' due to the mapping of DbType.DateTimeOffset (which does not have a matching NpgsqlDbType)

…etters # Conflicts: # src/Npgsql/NpgsqlParameter.cs

# Conflicts: # src/Npgsql/NpgsqlDataSourceBuilder.cs # src/Npgsql/NpgsqlSlimDataSourceBuilder.cs

This will allow custom resolvers to only return results under more specific circumstances

… implementation compatible with unqualified names

PauloHMattos requested review from roji and vonzshik as code owners October 25, 2025 02:26

NinoFloris reviewed Oct 25, 2025

View reviewed changes

inoa-pmattos force-pushed the db-type-resolver branch from 1ba1424 to f77aeff Compare October 27, 2025 18:09

NinoFloris reviewed Nov 8, 2025

View reviewed changes

src/Npgsql/NpgsqlParameter.cs Show resolved Hide resolved

test/Npgsql.Tests/TypeMapperTests.cs Outdated Show resolved Hide resolved

src/Npgsql/NpgsqlParameter.cs Outdated Show resolved Hide resolved

src/Npgsql/NpgsqlParameter.cs Outdated Show resolved Hide resolved

PauloHMattos commented Nov 8, 2025

View reviewed changes

NinoFloris reviewed Nov 9, 2025

View reviewed changes

src/Npgsql/NpgsqlParameter.cs Outdated Show resolved Hide resolved

src/Npgsql/NpgsqlParameter.cs Outdated Show resolved Hide resolved

src/Npgsql/NpgsqlParameter.cs Outdated Show resolved Hide resolved

inoa-pmattos force-pushed the db-type-resolver branch from 41d13f0 to 492f658 Compare November 10, 2025 13:12

NinoFloris force-pushed the db-type-resolver branch 2 times, most recently from fdbbe67 to 563831e Compare November 10, 2025 17:37

NinoFloris approved these changes Nov 10, 2025

View reviewed changes

NinoFloris force-pushed the db-type-resolver branch from 563831e to 75d2389 Compare November 10, 2025 17:39

NinoFloris mentioned this pull request Nov 10, 2025

Move all reloadable state into one reference #6303

Merged

NinoFloris force-pushed the db-type-resolver branch from 75d2389 to 0151b80 Compare November 10, 2025 21:34

This comment was marked as outdated.

Sign in to view

NinoFloris force-pushed the db-type-resolver branch from 0151b80 to c345c91 Compare November 10, 2025 21:40

This was referenced Nov 11, 2025

[Tests] Remove npgsqldbtype passing and rename related flag #6305

Merged

[Tests] Fully rework type mapping assertion helpers #6314

Merged

NinoFloris force-pushed the db-type-resolver branch 5 times, most recently from c4ae65c to f49844f Compare November 16, 2025 19:14

NinoFloris reviewed Nov 16, 2025

View reviewed changes

NinoFloris changed the title ~~Add IDbTypeResolver to allow PgTypeInfoResolverFactory to control how DbTypes are mapped~~ Add IDbTypeResolver to allow plugins to control how DbTypes are mapped Nov 16, 2025

roji reviewed Nov 18, 2025

View reviewed changes

src/Npgsql/NpgsqlDataSourceBuilder.cs Outdated Show resolved Hide resolved

Add IDbTypeResolver to allow PgTypeInfoResolverFactory to control how…

5e10a27

… DbTypes are mapped

NinoFloris added 14 commits November 18, 2025 19:35

Fix command builder issue

41ca81f

Adapt tests to having dedicated db type storage

68b7f79

Correctly handle DbType <-> NpgsqlDbType discontinuities

dbc4592

This showed up in 'Timestamptz_as_DateTimeOffset_utc_with_DbType_DateTimeOffset' due to the mapping of DbType.DateTimeOffset (which does not have a matching NpgsqlDbType)

Fix incorrect flag passed

4399dba

Only attempt to map the first non-null data type alternative inside g…

6e33eea

…etters # Conflicts: # src/Npgsql/NpgsqlParameter.cs

Style fixes

a2e1d56

Make DbTypeResolverFactory its own thing

74924ce

# Conflicts: # src/Npgsql/NpgsqlDataSourceBuilder.cs # src/Npgsql/NpgsqlSlimDataSourceBuilder.cs

Enable value inference on null generic parameters

24963e3

Make ResolveTypeInfo read more naturally

2fd26ac

Optionally pass any available value type to db type resolver

3486ec0

This will allow custom resolvers to only return results under more specific circumstances

Fix fall back during resolve

aba06f8

Bring back error on unsupported db type cases

36a6225

Tweak error message

b740d19

Move the new api surface to INpgsqlTypeMapper

18fc366

NinoFloris force-pushed the db-type-resolver branch 2 times, most recently from a1419af to 4d64c2a Compare November 18, 2025 18:53

Bring back the DataTypeName typed argument by making the DataTypeName…

208ffe6

… implementation compatible with unqualified names

NinoFloris force-pushed the db-type-resolver branch from 4d64c2a to 208ffe6 Compare November 18, 2025 18:54

NinoFloris merged commit b16f722 into npgsql:main Nov 18, 2025
13 checks passed

		Assert.That(parameter.NpgsqlDbType, Is.EqualTo(NpgsqlDbType.Uuid));
		Assert.That(parameter.DataTypeName, Is.EqualTo("uuid"));

Conversation

PauloHMattos commented Oct 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

roji commented Oct 25, 2025

Uh oh!

NinoFloris left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

PauloHMattos Nov 8, 2025

Choose a reason for hiding this comment

Uh oh!

NinoFloris Nov 8, 2025

Choose a reason for hiding this comment

Uh oh!

PauloHMattos Nov 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

PauloHMattos Nov 10, 2025

Choose a reason for hiding this comment

Uh oh!

NinoFloris Nov 10, 2025

Choose a reason for hiding this comment

Uh oh!

PauloHMattos Nov 10, 2025

Choose a reason for hiding this comment

Uh oh!

NinoFloris Nov 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

This comment was marked as outdated.

PauloHMattos commented Nov 12, 2025

Uh oh!

NinoFloris commented Nov 14, 2025

Uh oh!

NinoFloris Nov 16, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

PauloHMattos commented Oct 25, 2025 •

edited

Loading

PauloHMattos Nov 10, 2025 •

edited

Loading

NinoFloris Nov 10, 2025 •

edited

Loading