Figuring Out Foreach
For a while I have been aware of some odd behaviour when you use branch coverage with PHP's foreach language construct. Usually with PHPUnit.
Take for example this bit of code:
<?php
function showLanguages(array $languages): void
{
foreach ($languages as $language) {
echo $language, "\n";
}
}
It seems that in order to cover all paths by test cases to pass in an array without any elements, and one with at least one element.
But no, when you run the analysis, you will see that there only two out of the three paths are covered:
Let's first actually analyse which opcodes, PHP's internal execution units, are generated by the engine:
line #* E I O op return operands
------------------------------------------------
2 0 E > RECV !0
4 1 EXT_STMT
2 > FE_RESET_R $2 !0, ->9
3 > > FE_FETCH_R $2, !1, ->9
5 4 > EXT_STMT
5 ECHO !1
6 EXT_STMT
7 ECHO '%0A'
4 8 > JMP ->3
9 > FE_FREE $2
7 10 EXT_STMT
11 > RETURN null
This output from VLD shows on line 4, two relevant opcodes: FE_RESET_R and FE_FETCH_R. Both of them can either followed by the next opcode on the list, or they can jump (->9) to opcode 9, which marks the end of the loop.
However from the diagram above, we don't see that path (the green one) being taken. Only the two paths that both continue with the next opcode (the red and blue lines). It seems that the emptiness of a normal array is checked by the FE_FETCH_R opcode.
Now the question is how to trigger the other path, so that 100% path (and branch) coverage can be reached.
This turns out harder than it is. I originally thought that this would be possible by trying to create a broken iterator — for example one where you inherit from an internal class with a custom one, and not call the original constructor. But this creates an exception which pre-empts the engine from even running the rest of the function.
The only situation where I managed to do this was by creating an iterator that after a correct initialisation, had no items to return. An example of such as case is a DatePeriod iterator where the start date of the iterator is behind the end date:
$i1 = new DatePeriod(
new DateTimeImmutable("2025-01-14"),
DateInterval::createFromDateString("+1 day"),
new DateTimeImmutable("2025-01-01")
);
I had to change the definition of showLanguages too so that it accepts DatePeriod besides just array. But with that done, this specific iterator now lights up the green path:
There is currently an open bug in Xdebug's issue tracker to merge the branch analysis information for the two subsequent opcodes (FE_RESET_R and FE_FETCH_R).
I think I will now rather hide the jump away from the FE_RESET_R (and FE_RESET_RW) opcode.
But at last I now have an explanation as to where this phantom path came from.
Life Line
I've just put the first event in my calendar for... 2027.
Looking forwards to seeing a match in the snooker Masters at Ally Pally!
Under the Waterloo station arches at Leake Street.
This specific area is set aside for grafity artists.
RE: https://phpc.social/@phpc_tv/115917245540319542
This means you can now follow @xdebug from Mastodon too!
Created a bar and an events_venue; Deleted a pub
Updated a restaurant; Deleted a clothes shop
I walked 8.5km in 1h22m28s
I walked 5.0km in 47m19s
Added Trogolo, and fixed duplicated addresses
I've finished my first book of the year, The Basic Soldering Guide Handbook.
Now I "just" need to put the learned knowledge into practise.
I walked 1.8km in 19m57s
Merged pull request #1058
Sort Xdebug modes in particular order, change performance label
Calculate and print performance change
I walked 6.0km in 1h0m58s
I walked 1.1km in 10m19s
In times like this, it's actually fairly useful to be able to read a fair amount of Danish.
Merge branch 'v2022'
Go with 2022.16
Merge branch 'v2022'
Merge branch 'v2022'
Go with 2022.15
Do a shallow clone
Merge branch 'v2022'




Shortlink
This article has a short URL available: https://drck.me/fidgety-forach-jbc