Add mode for verifying internal function param defaults

[nikic]

A problem here is that the extra parameters show up in exception stack traces, and make tests spuriously fail.

[nikic]

cc @kocsismate

Ugh internal functions are so weird.

[kocsismate]

@nikic OMG, all these failures are due to bad default values? 😱I'll try to have a look at them

[cmb69]

Fixing the session_set_save_handler() stub should fix about half of the test failures. Either switch all default values to UNKNOWN, or have the second parameter variadic (which actually is more like what ZPP does).

[cmb69]

 ext/session/session.stub.php | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/ext/session/session.stub.php b/ext/session/session.stub.php
index 09c35aae28..8c13281243 100644
--- a/ext/session/session.stub.php
+++ b/ext/session/session.stub.php
@@ -37,7 +37,7 @@ function session_register_shutdown(): void {}
 /** @alias session_write_close */
 function session_commit(): bool {}
 
-function session_set_save_handler($open, $close = null, $read = null, $write = null, $destroy = null, $gc = null, $create_sid = null, $validate_sid = null, $update_timestamp = null): bool {}
+function session_set_save_handler($open, $close = UNKNOWN, $read = UNKNOWN, $write = UNKNOWN, $destroy = UNKNOWN, $gc = UNKNOWN, $create_sid = UNKNOWN, $validate_sid = UNKNOWN, $update_timestamp = UNKNOWN): bool {}
 
 function session_cache_limiter(string $cache_limiter = UNKNOWN): string|false {}
 
@@ -45,7 +45,7 @@ function session_cache_expire(?int $new_cache_expire = null): int|false {}
 
 function session_set_cookie_params($lifetime_or_options, string $path = UNKNOWN, string $domain  = "", ?bool $secure = null, ?bool $httponly = null): bool {}
 
-function session_start(array $options = []): bool {}
+function session_start(array $options = UNKNOWN): bool {}
 
 interface SessionHandlerInterface
 {

fixes all ext/session test failures for me.

[nikic]

@cmb69 The session_set_save_handler change looks good to me, but why the session_start change?

[kocsismate]

@nikic I've fixed quite some issues. Please, rebase when you'll have time so that I can see the remaining issues more easily :)

[kocsismate]

@nikic If all the wishes would come true this fast... :D

[cmb69]

The session_set_save_handler change looks good to me, but why the session_start change?

Oh, indeed, the signature is correct; however, session_set_save_handler_sid_002.phpt and two other tests fail on AppVeyor and locally for me because of a slightly different stack trace. Why is the argument shown on Windows, but not on Linux?

[kocsismate]

Am I right that the default value of the $errno and the $errstr parameters of stream_socket_client() and stream_socket_server() shouldn't be null (because ZVAL null != null)?

[cmb69]

@kocsismate, yes.

[kocsismate]

As far as I see there will lots of issues with null default values of non-nullable parameters (at least ext/filter had quite some), so I think starting the overhaul of nullable parameters has just become very necessary. :/

[kocsismate]

@nikic Can you please rebase this next week? It's not urgent at all, but we should have a look at the current state soon :)

[nikic]

A few next weeks later, this is now rebased :) I've already fixed a couple of issues that have cropped up.