Curious, I thought I had replied to this one already.

I can not reproduce this.

When I run this on Linux with strace, I just get this:

socket(AF_INET, SOCK_STREAM, IPPROTO_TCP) = 3
fcntl(3, F_SETFL, O_RDONLY|O_NONBLOCK) = 0
connect(3, {sa_family=AF_INET, sin_port=htons(9000), sin_addr=inet_addr("127.0.0.1")}, 16) = -1 EINPROGRESS (Operation now in progress)
poll([{fd=3, events=POLLIN|POLLPRI|POLLOUT}], 1, 200) = 1 ([{fd=3, revents=POLLIN|POLLOUT|POLLERR|POLLHUP}])
close(3)

And nothing else — perhaps it is different on OSX. Can you make something like an strace for OSX? I think it's called dtruss there.

I've taken a closer look and the setup is a bit more complicated than simply osx (because docker). I've managed to track the issue down to automatically connecting to an incorrect IP address, which turns out is the same IP address where a php-fpm instance is running on port 9000.

This apparently does not like to chat with dbgp, since php-fpm drops the connection after the initial xml packet, but this isn't picked up by xdebug.

This is a backtrace of what's happening, with some small remarks

[pid 9] <... accept resumed> {sa_family=AF_INET6, sin6_port=htons(56418), inet_pton(AF_INET6, "::ffff:172.18.0.13", &sin6_addr), sin6_flowinfo=0, sin6_scope_id=0}, [28]) = 5
[pid 9] times({tms_utime=725, tms_stime=398, tms_cutime=0, tms_cstime=0}) = 4308070207
[pid 9] poll([{fd=5, events=POLLIN}], 1, 5000) = 1 ([{fd=5, revents=POLLIN}])
[pid 9] read(5, "\1\1\0\1\0\10\0\0", 8) = 8
[pid 9] read(5, "\0\1\0\0\0\0\0\0", 8) = 8
[pid 9] read(5, "\1\4\0\1\2J\6\0", 8) = 8
[pid 9] read(5, "\f\0QUERY_STRING\16\3REQUEST_METHODGE"..., 592) = 592
[pid 9] read(5, "\1\4\0\1\0\0\0\0", 8) = 8
[pid 9] lstat("/var/www/apipoc/web/app.php", {st_mode=S_IFREG|0664, st_size=53, ...}) = 0
[pid 9] lstat("/var/www/apipoc/web", {st_mode=S_IFDIR|0775, st_size=306, ...}) = 0
[pid 9] lstat("/var/www/apipoc", {st_mode=S_IFDIR|0755, st_size=748, ...}) = 0
[pid 9] lstat("/var/www", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
[pid 9] lstat("/var", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
[pid 9] stat("/var/www/apipoc/web/.user.ini", 0x7ffffd99a010) = -1 ENOENT (No such file or directory)
[pid 9] setitimer(ITIMER_PROF, {it_interval={0, 0}, it_value={30, 0}}, NULL) = 0
[pid 9] rt_sigaction(SIGPROF, {0x55be0fe57940, [PROF], SA_RESTORER|SA_RESTART, 0x7f90d83ee0e0}, {0x55be0fe57940, [PROF], SA_RESTORER|SA_RESTART, 0x7f90d83ee0e0}, 8) = 0
[pid 9] rt_sigprocmask(SIG_UNBLOCK, [PROF], NULL, 8) = 0
[pid 9] getcwd("/var/www/apipoc", 4095) = 16
[pid 9] chdir("/var/www/apipoc/web") = 0
[pid 9] fcntl(4, F_SETLK, {type=F_RDLCK, whence=SEEK_SET, start=1, len=1}) = 0
[pid 9] stat("/var/www/apipoc/web/app.php", {st_mode=S_IFREG|0664, st_size=53, ...}) = 0

opening Xdebug connection to 172.18.0.3:9000 (which runs php-fpm, not an xdebug listener)

[pid 9] socket(PF_INET, SOCK_STREAM, IPPROTO_TCP) = 7
[pid 9] fcntl(7, F_SETFL, O_RDONLY|O_NONBLOCK) = 0
[pid 9] connect(7, {sa_family=AF_INET, sin_port=htons(9000), sin_addr=inet_addr("172.18.0.3")}, 16) = -1 EINPROGRESS (Operation now in progress)
[pid 9] poll([{fd=7, events=POLLIN|POLLPRI|POLLOUT}], 1, 200) = 1 ([{fd=7, revents=POLLOUT}])
[pid 9] getpeername(7, {sa_family=AF_INET, sin_port=htons(9000), sin_addr=inet_addr("172.18.0.3")}, [16]) = 0
[pid 9] fcntl(7, F_SETFL, O_RDONLY) = 0
[pid 9] setsockopt(7, SOL_TCP, TCP_NODELAY, "\1\0\0\0\0\0\0\0", 8) = 0
[pid 9] write(7, "493\0<?xml version=\"1.0\" encoding"..., 498) = 498

php-fpm closes connection

[pid 9] recvfrom(7, 0x7ffffd998cb0, 128, 0, 0, 0) = -1 ECONNRESET (Connection reset by peer)
[pid 9] setitimer(ITIMER_PROF, {it_interval={0, 0}, it_value={0, 0}}, NULL) = 0
[pid 9] read(10, "\320\356 n\303\264\r+", 8) = 8
[pid 9] chdir("/var/www/apipoc") = 0
[pid 9] times({tms_utime=725, tms_stime=398, tms_cutime=0, tms_cstime=0}) = 4308070208
[pid 9] stat("/etc/localtime", {st_mode=S_IFREG|0644, st_size=118, ...}) = 0
[pid 9] write(6, "172.18.0.13 - 01/Mar/2017:10:48"..., 61) = 61
[pid 9] setitimer(ITIMER_PROF, {it_interval={0, 0}, it_value={30, 0}}, NULL) = 0

xdebug didn't handle closed connection and happily sends more data

[pid 9] write(7, "173\0<?xml version=\"1.0\" encoding"..., 178) = -1 EPIPE (Broken pipe)
[pid 9] --- SIGPIPE {si_signo=SIGPIPE, si_code=SI_USER, si_pid=9, si_uid=33} ---
[pid 9] write(2, "There was a problem sending 178 "..., 49) = 49
[pid 9] recvfrom(7, "", 128, 0, NULL, NULL) = 0
[pid 9] close(7) = 0

[pid 9] fcntl(4, F_SETLK, {type=F_UNLCK, whence=SEEK_SET, start=0, len=0}) = 0
[pid 9] write(5, "\1\6\0\1\0T\4\0X-Powered-By: PHP/7.0.15"..., 112) = 112
[pid 9] shutdown(5, SHUT_WR) = 0
[pid 9] recvfrom(5, "\1\5\0\1\0\0\0\0", 8, 0, NULL, NULL) = 8
[pid 9] recvfrom(5, "", 8, 0, NULL, NULL) = 0
[pid 9] close(5) = 0
[pid 9] munmap(0x7f90c9a00000, 2097152) = 0
[pid 9] setitimer(ITIMER_PROF, {it_interval={0, 0}, it_value={0, 0}}, NULL) = 0
[pid 9] accept(0,

I'll improve the error message, but it's working as expected:

Xdebug can't detect that a connection has been closed until it tries to write again. And when it tries to write, the write fails and Xdebug shows an error message.

In this case, the error message rightfully pointed out that something was wrong.

Improved the error message for 2.6dev.

I noticed this happens in a specific instance for me. When phpstorm either has too many debug connections open or the connection has timed out. Closing php storm resolved the issue for me.