STATUS AND COMPATIBILITY

How much PHP does phpr actually run?

Coverage was fully measured against PHP 8.5.7 on 30 July 2026; the server lifecycle snapshot is updated through 31 July 2026.

2,649

Zend tests pass · 65.1% of runnable tests

539

core functions out of 654 · 82%

1,017

internal functions out of 2,143 · 47%

30,472

WordPress single-site tests · multisite 31,278

Core language stdlib

82%

Strings, arrays, math, variables, date, output, hashing and regex: the surface ordinary applications use most.

All internal functions

47%

The remaining gap is dominated by whole specialist extension families, not missing fundamentals of the language.

COMPLETE AREAS

Surfaces closed function by function.

ctype · 100%json · 100%SimpleXML · 100%zlib · 100%bcmath · 100%tokenizer · 100% functionssession · 23/23xml · 22/22fileinfo · 6/6tidy · 24/24PDO core · 100%

“Complete” means all expected internal functions for that area are present; upstream suites can still include a documented edge case or environment-bound condition.

REAL SOFTWARE

What runs end to end.

These results traverse many runtime surfaces at once and are guarded by named regression baselines.

Project	Measured status	What it validates
Composer	Runs	`require`, `install`, `diagnose`, HTTPS and autoloading
PHPUnit	9.6 / 11.5 / 13.x	CLI, output buffering, Reflection and process isolation
Doctrine DBAL	3,769 / 0 / 0	PDO, SQLite, exceptions, metadata and SQLSTATE
Doctrine ORM	3,484 tests · 3 errors · 13 failures	Hydration, UnitOfWork, lazy proxies and Reflection
Symfony HttpKernel	1,665 tests · 0 errors · 0 failures	DI container, kernel lifecycle, time zones, eval scope and destructors
WordPress	30,472 single-site · 31,278 multisite	One declared name-diff in each suite; real MySQL, HTTP, REST, wp-admin, media and permalinks

2.06–2.11×

current reference band for full-suite CPU vs the oracle

2.58×

WordPress media-group CPU vs the oracle

~3.0–3.1×

current media footprint band vs the oracle

0.000

measured objects retained per request after the WP-72 teardown fix

CURRENT LIMITS

Where phpr is not PHP yet.

The full extension surface is incomplete

Large remaining examples include PostgreSQL, sodium, LDAP, ODBC, sockets, FTP, XMLWriter and other specialist APIs. Class-only surfaces also make function counts an incomplete picture.

“Effective parity” includes one declared difference

Both complete WordPress suites differ by one named test because stream_get_wrappers() honestly lists only phpr's built-in wrappers. The divergence is catalogued rather than hidden behind a lying result.

The worker arena persists; the VM does not yet

Each dedicated Axum worker now retains one RetainSet across sequential requests, and the two-request opening gate is byte-identical. The integration still constructs a fresh VM and module per request; true same-VM reuse and amplification gates remain open.

No production promise

The server paths are real and heavily exercised, but security review, deployment packaging, multi-platform validation and sustained-load behavior are not yet sufficient to replace PHP in production.

Measured data: COVERAGE.md ↗ · performance trend: GAP_TREND.md ↗ · current route: NEXT_SESSION_WORDPRESS.md ↗ · known divergences ↗