The Mystery of the Missing Breakpoints
Occasionally I see people mentioned that Xdebug does not stop at certain breakpoints. This tends to relate to multi-line if conditions, or if/else conditions without braces ({ and }). Take for example the following artificial code sample:
1 <?php
2 $a = true;
3 $b = true;
4 $c = true;
5 $d = false;
6
7 if (
8 $a
9 && $b
10 && ( $c || $d )
11 )
12 {
13 echo "It's true!";
14 }
If you set a breakpoint at either line 7, 11, or 12, you'll find that these are ignored. But why is that?
In order to investigate we employ vld, a tool that I wrote that can show PHP's internal bytecode (oparrays containing opcodes). They read very much like assembly code. For the above snippet, the vld dump looks as follows (after redacting useless information):
function name: (null)
number of ops: 19
compiled vars: !0 = $a, !1 = $b, !2 = $c, !3 = $d
line #* E I O op return operands
---------------------------------------------------
2 0 E > EXT_STMT
1 ASSIGN !0, <true>
3 2 EXT_STMT
3 ASSIGN !1, <true>
4 4 EXT_STMT
5 ASSIGN !2, <true>
5 6 EXT_STMT
7 ASSIGN !3, <false>
8 8 EXT_STMT
9 > JMPZ_EX ~8 !0, ->11
9 10 > BOOL ~8 !1
11 > > JMPZ_EX ~8 ~8, ->15
10 12 > > JMPNZ_EX ~9 !2, ->14
13 > BOOL ~9 !3
14 > BOOL ~8 ~9
15 > > JMPZ ~8, ->18
13 16 > EXT_STMT
17 ECHO 'It%27s+true%21'
15 18 > > RETURN 1
The first column is the line number that PHP associates with each opcode, and you can see that there is no opcode for lines 7, 11, and 12. Additionally, Xdebug can only break on the EXT_STMT opcode, which you can see is only present for lines 8 and 13 in the logic section of the script. If a breakpoint is set on a line without an EXT_STMT opcode, Xdebug will not be able to interrupt the script on that line.
Sometimes it can be confusing where PHP thinks there is a line of code, especially since OPcache starts optimising more and more things out. In our example, it could really only leave line 13 around, as the rest is static. It doesn't quite do that yet however.
It is certainly frustrating that Xdebug cannot always stop where you want it to, but IDEs do have a possibility to interrogate where Xdebug could stop through a private DBGp commandxcmd_get_executable_lines. This can only be done when the script, or rather, a specific function is already running.
The DBGp protocol has provisions for signalling to IDEs whether it has resolved a breakpoint when a function gets entered into. Xdebug does currently not implement this functionality yet, but a ticket for it is scheduled for implementation (likely for Xdebug 3.0).
Life Line
Updated a restaurant; Confirmed a hotel
Paraphrasing opening keynote speaker at ConFoo: "Should we go back to the waterfall method of writing massive specs upfront to feed to AI coding agents?"
Updated a pub
Merged pull request #1065
Comparison whether class is userland or internal used the wrong macro
PHP 8.6: zend_enum.h now mixes code with declarations
PHP 8.6: Argument names are now stored as zend_strings
Updated a bench and a waste_basket
I walked 8.3km in 1h25m37s
Created a recycling
I walked 10.5km in 1h46m57s
An interesting journey in story form, showing how English changed over time.
https://www.deadlanguagesociety.com/p/how-far-back-in-time-understand-english
A much better writer than I is summing up perfectly why I have such disdain for Generative AI/LLMs.
https://jonn.substack.com/p/so-why-do-i-feel-so-angry-about-this
Created a waste_basket; Updated a waste_basket; Deleted a bench
Created a bench; Updated 7 benches and a gate; Deleted 2 benches and a gate
Created 10 benches and 2 waste_baskets; Updated an information
Created a bench; Updated 2 benches; Deleted a bench
I hiked 18.1km in 3h17m10s
I walked 3.0km in 25m12s
Updated a restaurant
I walked 4.6km in 34m18s
I walked 0.7km in 5m33s
Updated a restaurant



Shortlink
This article has a short URL available: https://drck.me/brk-ea9