Improve build time of setting classes

[Algunenano]

Changelog category (leave one):

• Not for changelog (changelog entry is not required)

Changelog entry (a user-readable short description of the changes that goes into CHANGELOG.md):

Improve build time of setting classes

Introduce a base class SettingFieldBase for all SettingField##Type classes in order to remove most of the lambda functions in FieldInfo. This makes the classes larger (because of the virtual tables) but speeds up compilation, which is a pain.

I also tried to move from using one member per setting inside of ::Data to use a std::vector of SettingFieldBase but this made the code way more complex and slower to build (15->25 seconds).

Comparing Settings.cpp (the largest one) in my local dev machine:

• Before:

real    0m43.484s
user    0m42.696s
sys     0m0.548s

• After:

real    0m15.649s
user    0m15.095s
sys     0m0.307s

References #58797

Documentation entry for user-facing changes

• Documentation is written (mandatory for new features)

[clickhouse-gh]

Workflow [PR], commit [dfc2929]

Summary: ❌

job_name	test_name	status	info	comment
AST fuzzer (amd_ubsan)		failure
	Logical error: Cannot convert nested result of function A with type B to the expected result type C: D (STID: None)	FAIL	cidb	IGNORED
BuzzHouse (amd_debug)		failure
	Logical error: 'Inconsistent AST formatting: the query: (STID: 1941-1bfa)	FAIL	cidb, issue	ISSUE EXISTS
Stateless tests (arm_asan, targeted)		error		IGNORED

[Copilot]

Pull request overview

This PR introduces a base class SettingFieldBase for all SettingField* types to improve build time of setting classes by reducing the number of generated lambda functions. The change makes the setting field classes larger (due to virtual tables) but significantly speeds up compilation, reducing build time for Settings.cpp from ~43 seconds to ~15 seconds.

Changes:

• Introduced SettingFieldBase with virtual methods for common operations
• Made all SettingField* classes inherit from SettingFieldBase
• Simplified FieldInfo metadata to only store two function pointers instead of many lambdas
• Changed getPath() to return const char* and handle empty paths for custom settings

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 10 comments.

File	Description
src/Core/SettingsFields.h	Introduces SettingFieldBase base class and updates all SettingField* classes to inherit from it with virtual method overrides
src/Core/ServerSettings.cpp	Updates getPath() usage to handle empty strings for settings without paths
src/Core/BaseSettings.h	Refactors FieldInfo to use two function pointers instead of many lambdas, updates getPath() return type, adds extensive documentation

Comments suppressed due to low confidence (1)

src/Core/BaseSettings.h:1

• Multiple uses of const_cast throughout these methods. Consider adding a const-qualified version of get_data_function or redesigning to avoid casting away constness, which can lead to undefined behavior if the underlying data is truly const.

#pragma once

[Copilot]

Virtual destructor should use = default instead of empty braces {} to clearly indicate default behavior and allow compiler optimizations.

Suggested change

	virtual ~SettingFieldBase() {}
	virtual ~SettingFieldBase() = default;

[Copilot]

constexpr String is invalid in C++. Use static const String or static constexpr std::string_view instead.

[clickgapai]

Hi @Algunenano @nickitat — while reviewing this PR I found the following:

• Bug (P2): resetValueToDefault round-trip through Field loses SettingFieldMaxThreads::is_auto flag — resetValueToDefault round-trip through Field loses SettingFieldMaxThreads::is_auto flag #103120

Happy to discuss — close anything that's wrong or already addressed.