fix: don't force-overwrite argv if do_php_cli is called by embedder

[henderkes]

https://learn.microsoft.com/en-us/cpp/c-runtime-library/argc-argv-wargv?view=msvc-170

if we get called by an embedder (argv != __argv) don't re-parse the cmd.

[henderkes]

cc @alexandre-daubois as this concerns frankenphp

cc @shivammathur for windows-specific review I think

[shivammathur]

Can we test the Windows do_php_cli() argument contract, also I agree with Alexandre that the embedder-supplied narrow argv skips transcoding.

[henderkes]

Hmm, small issue in this is also that we're kind of expecting embedders to pass argv in the internal encoding (which is utf-8 by default, but short of explicitly overwriting it, an embedder couldn't be sure). Should we instead require utf-8 and do the conversion if necessary? It's a bit extra code.

Edit: went with always taking utf-8, I don't see another good way of dealing with it.

[henderkes]

@shivammathur @alexandre-daubois what do you think, pass argv as utf-8 or always as wchar_t and always do the encoding like before?

[shivammathur]

@henderkes

Thinking more about this, I think we can preserve argv and document the caller’s encoding responsibility. That’s simpler and avoids misinterpreting UTF-8 -d values when PHP uses Windows-1252 for example.

[henderkes]

I don't think that's a good idea, because the embedding program may not know php's internal encoding.

[shivammathur]

Yes, unless the embedder explicitly sets it.
If we go with UTF-8, we should handle cases like -d auto_prepend_file=café.php when default_charset=Windows-1252. Currently, that value is applied before conversion, so PHP looks for the wrong filename.

[henderkes]

Yes, unless the embedder explicitly sets it. If we go with UTF-8, we should handle cases like -d auto_prepend_file=café.php when default_charset=Windows-1252. Currently, that value is applied before conversion, so PHP looks for the wrong filename.

But cli never takes the UTF-8 path, it's still going into the argv == __argv branch and converts from wstring to string.