Catchable "Call to a member function bar() on a non-object" #647

thekid · 2014-04-19T12:15:06Z

This pull request turns fatal errors generated from calls to methods on a non-object into E_RECOVERABLE_ERRORs.

Example

set_error_handler(function($code, $message) {
  var_dump($code, $message);
});

$x= null;
var_dump($x->method());
echo "Alive\n";

The above produces the following output:

int(4096)
string(50) "Call to a member function method() on a non-object"
NULL
Alive

Reasoning

If you want to run PHP as a webserver itself, fatal errors are problematic. For a long story on why you would want to do that in the first place, see http://marcjschmidt.de/blog/2014/02/08/php-high-performance.html

Consistency

This behavior is consistent with how type hints work. Framework authors can turn this into exceptions if they wish.

Inner workings

The essence of this PR is a patch to the Zend Engine, which raises an E_RECOVERABLE_ERROR and then skips over all opcodes until it finds the one ending the current call, and sets return value to NULL.

See this gist for the patch to Zend/zend_vm_def.h only, the full diff view contains tests as well as generated code.

RFC: Engine exceptions - this PR doesn't go as far as this controversial RFC
PHP Bugs 46601, 51882 and 51848 - bugs which would be fixed by this
PHP Bug 54195 - related, motivates necessity
HHVM throws a BadMethodCallException in these situations - can be done by implementing an error handler as seen here

smalyshev · 2014-04-20T22:35:10Z

This is a very nicely made pull request. However, I would say this would require an RFC or at least discussion on internals list to see if we have support for it.

thekid · 2014-04-22T08:38:09Z

@smalyshev - thanks for the feedback! I will come up with an RFC ASAP.

thekid · 2014-04-26T14:43:02Z

First draft for an RFC @ https://wiki.php.net/rfc/catchable-call-to-member-of-non-object

thekid · 2014-04-27T12:24:51Z

Set the RFC to 1️⃣.0️⃣ and started discussion on internals mailing list.

http://news.php.net/php.internals/73823

thekid · 2014-06-29T10:53:37Z

Set to voting status @ https://wiki.php.net/rfc/catchable-call-to-member-of-non-object#vote

nikic · 2014-07-04T13:51:16Z

Zend/zend_vm_def.h

+		EX_T(opline->result.var).var.fcall_returned_reference = 0;
+		EX_T(opline->result.var).var.ptr_ptr = &EX_T(opline->result.var).var.ptr;
+		ZEND_VM_NEXT_OPCODE();
+		return;


This return; will break the goto and switch VMs. You can just drop it, as ZEND_VM_NEXT_OPCODE() will already transfer control from here.

nikic · 2014-07-04T14:32:13Z

Another thing to consider is how this interacts with EXT_FCALL_BEGIN/END. As currently implemented, I think this will result in EXT_FCALL_BEGIN to be skipped, but EXT_FCALL_END being run, which doesn't sounds right. You can test this using sapi/cli/php -E, if I remember correctly.

From the tests side, It would be nice to test calls with non-cv operands, e.g. $arr[0]->{$arr[1]}(), as well as nested calls.

thekid · 2014-07-04T16:05:51Z

@nikic - thanks for the detailled feedback! I'll look into this ASAP and give you feedback one by one.

@nikic

See feedback by @nikic, php#647 (comment)

See php#647 (comment)

See discussion php#647 (comment)

@nikic

Verified with running tests with new "-e" run-tests arg: $ make test TESTS=Zend/tests/*-on-non-objects-*phpt TEST_PHP_ARGS=-e # Tests passed : 11 (100.0%) $ make test TESTS=Zend/tests/*-on-non-objects-*phpt # Tests passed : 11 (100.0%) Before, this would cause a SEGV. Thanks @nikic for raising this concern

…thekid/php-src into catchable-fatals/methods-on-non-objects

nikic · 2014-07-09T15:27:14Z

Zend/zend_vm_execute.h

+		 * nesting level. Return NULL (except when return value unused) */
+		do {
+			ZEND_VM_INC_OPCODE();
+			opline++;


This won't work correctly with switch/goto VMs, because both lines will increment the opline variable. I would suggest to instead only increment opline here (and also on line 9347) and replace the CHECK_EXCEPTION() below with SAVE_OPLINE(). (Actually check_exception and save_opline are the same, but save_opline is probably more clear about what you want here)

Using the following didn't work for me:

diff --git a/Zend/zend_vm_def.h b/Zend/zend_vm_def.h index 42d3f60..628db06 100644 --- a/Zend/zend_vm_def.h +++ b/Zend/zend_vm_def.h @@ -2477,7 +2477,6 @@ ZEND_VM_HANDLER(112, ZEND_INIT_METHOD_CALL, TMP|VAR|UNUSED|CV, CONST|TMP /* No exception raised: Skip over arguments until fcall opcode with correct * nesting level. Return NULL (except when return value unused) */ do { - ZEND_VM_INC_OPCODE(); opline++; } while (opline->opcode == ZEND_DO_FCALL_BY_NAME ? opline->op2.num > nesting : @@ -2488,10 +2487,10 @@ ZEND_VM_HANDLER(112, ZEND_INIT_METHOD_CALL, TMP|VAR|UNUSED|CV, CONST|T } if ((opline + 1)->opcode == ZEND_EXT_FCALL_END) { - ZEND_VM_INC_OPCODE(); + opline++; } - CHECK_EXCEPTION(); + SAVE_OPLINE(); ZEND_VM_NEXT_OPCODE(); }

The SAVE_OPLINE() macro only sets EX(opline) whereas ZEND_VM_INC_OPCODE(); affects OPLINE which is authorative for the executor AFAIS.

Seems ZEND_VM_SET_OPCODE() might be what I'm looking for.

Yeah, sorry, I forgot that SAVE_OPLINE only sets EX(opline) for GOTO/SWITCH, but is a noop for CALL. ZEND_VM_SET_OPCODE sounds more like it.

nikic · 2014-07-09T15:31:15Z

@thekid Thanks for updating everything so quickly! I've added two last notes, though one is purely cosmetic, so feel free to ignore it ;)

…ODE() calls

thekid · 2014-07-31T18:07:39Z

This RFC has now been accepted! I've updated the RFC to reflect this.

Last step: Merge it. @nikic - can you do this? I don't have karma for Zend/...

with `zend_do_convert_call_user_func()` in Zend/zend_compile.c

thekid · 2014-08-16T15:47:43Z

I've brought this pull request up-to-date so it's once again mergeable with master. I had to resort back to counting ZEND_INIT_*CALL* and ZEND_DO_FCALL opcodes for handling nesting correctly again since the nesting level no longer is in op2.num. The code now uses the same list of opcodes as in zend_do_convert_call_user_func() in Zend/zend_compile.c...

$ make test TESTS=Zend/tests/*methods-*phpt
# [...]
=====================================================================
Number of tests :   16                16
Tests skipped   :    0 (  0.0%) --------
Tests warned    :    0 (  0.0%) (  0.0%)
Tests failed    :    0 (  0.0%) (  0.0%)
Expected fail   :    0 (  0.0%) (  0.0%)
Tests passed    :   16 (100.0%) (100.0%)
---------------------------------------------------------------------
Time taken      :    1 seconds
=====================================================================

See a1a4ba9#commitcomment-7414362

See a1a4ba9#commitcomment-7414223

thekid · 2014-08-29T12:23:40Z

From https://www.mail-archive.com/[email protected]/msg68933.html

On Aug 16, 2014, at 9:09, Timm Friebe [email protected] wrote:
two weeks ago, the RFC Catchable "Call to a member function bar() on a non-object" was accepted by a vote. I don't have commit access to
php-src/Zend,

If it's not sorted by someone else, I land this when I get back home later
today.

@sgolemon any news?

datibbaw · 2014-09-15T22:20:47Z

Could you squash the branch into a single commit and edit the NEWS and UPGRADING files?

thekid · 2014-09-16T19:48:23Z

I'm off to a conference for the next couple of days and will look into getting this done ASAP after I get back; the earliest next Sunday

thekid · 2014-09-27T18:19:44Z

Could you squash the branch into a single commit

No matter how hard I try git rebase always fails somewhere in the middle. I'll create a new PR with a single commit.

Changed ZEND_INIT_METHOD_CALL to raise an E_RECOVERABLE_ERROR and handle error handler outcome, returning NULL if error handler doesn't terminate the script. Adjusted outcome in MySQL / PDO_MySQL and ext/date Added various examples and tests Added tests for indirect calls - see http://news.php.net/php.internals/73823 Added tests verifying method calls on non-objects work inside eval() Originally developed inside php#647

hikari-no-yume · 2014-10-04T23:35:24Z

Is this going to be merged?

thekid · 2014-10-05T12:09:49Z

@TazeTSchnitzel I' m trying, but everytime it is a mergeable state there's noone around to do so. I myself don't have commit access to the upstream project...

thekid · 2014-10-05T12:34:21Z

Closing this, #847 is the new place

* pr/647: (33 commits) zend_uint -> uint32_t Fix nesting for *non*-compile-time-resolveable functions See thekid@a1a4ba9#commitcomment-7414223 Add tests for calls to nested, *non*-compile-time-resolveable functions See thekid@a1a4ba9#commitcomment-7414362 Make list of opcodes used for nesting calculation consistent with `zend_do_convert_call_user_func()` in Zend/zend_compile.c Rewrite code to use ZEND_VM_JMP() instead of repeated ZEND_VM_INC_OPCODE() calls QA: Simplify code to find matching ZEND_DO_FCALL_BY_NAME CG(context).nested_calls is stored inside the initializer's result.num and inside the finalizer's op2.num, by comparing these we don't need to count manually, and are thus safer from future expansion with specialized opcodes e.g. Fix expected fatal error, now is catchable fatal Adjust expected fatal error message Now also includes "on [TYPE]" after merge from master Check for memory leaks when not using return value Adjust expected fatal error message Now also includes "on [TYPE]" after merge from master Add tests with arrays as parameters Handle ZEND_NEW nesting Also verify nesting with dynamically called static methods Handle ZEND_INIT_NS_FCALL_BY_NAME nesting QA: Refactor: Split tests a bit to make them more comprehendable Support nested static calls Handle ZEND_EXT_FCALL_END, skipping if necessary Verified with running tests with new "-e" run-tests arg: $ make test TESTS=Zend/tests/*-on-non-objects-*phpt TEST_PHP_ARGS=-e # Tests passed : 11 (100.0%) Add support for PHP's 'extended information for debugger/profiler' mode Verify non-CV-operands also work See discussion #647 (comment) Only allocate NULL return value if it's actually used ... Conflicts: ext/date/tests/bug67118.phpt

datibbaw · 2014-10-05T22:00:53Z

Oh shit, the numbers looked so alike that I had merged in this PR instead of the newer one.

They're the same changes minus the added test cases, so instead of reverting what I have already done I shall just add the additional test cases.

My bad =(

thekid · 2014-10-05T23:10:30Z

I [...] merged in this PR

\o/

thekid added 7 commits April 19, 2014 14:01

Initial implementation

bdb0cee

Adjust expected outcome

daf16d5

Regenerate

f5efee1

Adjust expected outcome (non-Zend tests)

9265077

Make E_RECOVERABLE_ERRORs (4096) show up

a6f7f58

Add example how framework code could convert E_RECOVERABLE -> exceptions

51a5e18

Add test verifying chained method invocations

a86e762

thekid added 2 commits April 28, 2014 21:44

Add tests for indirect calls to methods on non-objects

da1db5e

http://news.php.net/php.internals/73823

Add tests verifying method calls on non-objects work inside eval()

7041ef4

nikic reviewed Jul 4, 2014
View reviewed changes

thekid added 10 commits July 6, 2014 12:27

Handle nesting correctly

17b3791

See feedback by @nikic, php#647 (comment)

Remove return statement which would break goto and switch VMs

2758824

See php#647 (comment)

QA: Simplify code to initialize a NULL value

abd42ea

See php#647 (comment)

Also handle nested function calls via variable

a782f69

Only allocate NULL return value if it's actually used

dc68aa3

Verify non-CV-operands also work

e0cf63f

See discussion php#647 (comment)

Add support for PHP's 'extended information for debugger/profiler' mode

ab4ded9

Support nested static calls

ff66e59

QA: Refactor: Split tests a bit to make them more comprehendable

7cd8ecb

Merge branch 'catchable-fatals/methods-on-non-objects' of github.com:…

a9653b6

…thekid/php-src into catchable-fatals/methods-on-non-objects

nikic reviewed Jul 9, 2014
View reviewed changes

Rewrite code to use ZEND_VM_JMP() instead of repeated ZEND_VM_INC_OPC…

a49ca6f

…ODE() calls

thekid added 2 commits August 16, 2014 17:38

MFH

c3d214a

Make list of opcodes used for nesting calculation consistent

a1a4ba9

with `zend_do_convert_call_user_func()` in Zend/zend_compile.c

thekid added 2 commits August 16, 2014 22:03

Add tests for calls to nested, *non*-compile-time-resolveable functions

971d369

See a1a4ba9#commitcomment-7414362

Fix nesting for *non*-compile-time-resolveable functions

3db6731

See a1a4ba9#commitcomment-7414223

thekid and others added 3 commits September 25, 2014 19:45

Merge branch 'master' into catchable-fatals/methods-on-non-objects

e83580c

zend_uint -> uint32_t

0a9e207

Merge branch 'master' into catchable-fatals/methods-on-non-objects

5c6ca24

thekid mentioned this pull request Sep 27, 2014

Implement the RFC Catchable "Call to a member function bar()..." #847

Closed

thekid closed this Oct 5, 2014

thekid deleted the catchable-fatals/methods-on-non-objects branch October 5, 2014 23:23

Catchable "Call to a member function bar() on a non-object" #647

Catchable "Call to a member function bar() on a non-object" #647

Uh oh!

Conversation

thekid commented Apr 19, 2014

Example

Reasoning

Consistency

Inner workings

Further reading

Uh oh!

smalyshev commented Apr 20, 2014

Uh oh!

thekid commented Apr 22, 2014

Uh oh!

thekid commented Apr 26, 2014

Uh oh!

thekid commented Apr 27, 2014

Uh oh!

thekid commented Jun 29, 2014

Uh oh!

nikic Jul 4, 2014

Choose a reason for hiding this comment

Uh oh!

nikic commented Jul 4, 2014

Uh oh!

thekid commented Jul 4, 2014

Uh oh!

nikic Jul 9, 2014

Choose a reason for hiding this comment

Uh oh!

thekid Jul 12, 2014

Choose a reason for hiding this comment

Uh oh!

nikic Jul 12, 2014

Choose a reason for hiding this comment

Uh oh!

nikic commented Jul 9, 2014

Uh oh!

thekid commented Jul 31, 2014

Uh oh!

thekid commented Aug 16, 2014

Uh oh!

thekid commented Aug 29, 2014

Uh oh!

datibbaw commented Sep 15, 2014

Uh oh!

thekid commented Sep 16, 2014

Uh oh!

thekid commented Sep 27, 2014

Uh oh!

hikari-no-yume commented Oct 4, 2014

Uh oh!

thekid commented Oct 5, 2014

Uh oh!

thekid commented Oct 5, 2014

Uh oh!

datibbaw commented Oct 5, 2014

Uh oh!

thekid commented Oct 5, 2014

Uh oh!

Uh oh!