declare(strict_types=1) does not ignore null AST nodes

[FearXenon]

Description

The following code:

<?php
// e.g some comments
?>
<?php
declare(strict_types=1);
?>

Resulted in this output:

Fatal error: strict_types declaration must be the very first statement in the script in /home/tmp/test.php on line 5
Stack trace:
#0 {main}

But I expected this output instead:

PHP Version

nightly

Operating System

No response

[devnexen]

Hi @FearXenon what s the practical reason for not doing this ?

<?php
// e.g some comments

declare(strict_types=1);
?>

AFAIK, it might be possible to "fix" it, but do not know how trivial it is. cc @iluuu1994

[iluuu1994]

Not really a bug. ?> includes an implicit ;, so it's not actually the first statement.

[FearXenon]

Hi @devnexen,
i´ve seen some projects that use a separate script to prepend licensing/copyright info to php files.
some of them did stuff like this:

<?php
Copyright (c) 2025 John Doe
...
?>
<?php
//start of normal code
?>

besides, I don´t really see the point of why it shouldn´t work, since technically nothing gets "executed" before, right?

[devnexen]

I see, but again that seems a very particular case. "How much the juice is worth the squeeze" is the question. But I ll let the lexer expert speak :)

[ndossche]

We do allow it for namespaces, so why not also for strict_types? I don't see a reason why this wouldn't work (I could be missing some detail):

diff --git a/Zend/zend_compile.c b/Zend/zend_compile.c
index e01b985b6d6..90d305dba34 100644
--- a/Zend/zend_compile.c
+++ b/Zend/zend_compile.c
@@ -7004,7 +7004,7 @@ static void zend_compile_declare(zend_ast *ast) /* {{{ */
 		} else if (zend_string_equals_literal_ci(name, "strict_types")) {
 			zval value_zv;
 
-			if (FAILURE == zend_is_first_statement(ast, /* allow_nop */ 0)) {
+			if (FAILURE == zend_is_first_statement(ast, /* allow_nop */ true)) {
 				zend_error_noreturn(E_COMPILE_ERROR, "strict_types declaration must be "
 					"the very first statement in the script");
 			}

[iluuu1994]

@nielsdos Should be ok, but would handle it as a feature rather than bug. Not sure if the same goes for declare(encoding=);.

[ndossche]

@nielsdos Should be ok, but would handle it as a feature rather than bug. Not sure if the same goes for declare(encoding=);.

Yeah sure, I can submit it as a new feature (for after 8.5). I don't think we should do encoding as that may have influence on some special characters that may have influence.