workaround for a bug in sysread

There is a bug in sysread from zsh/system. It triggers in the
following case:

1. zsh has been compiled with HAVE_SELECT and without HAVE_POLL.
2. sysread is called with timeout (-t).
3. the input file descriptor is valid but there is no data to read.
4. errno happens to be EINTR prior to the call to sysread.

This results in an infinite loop in sysread:

  while ((ret = select(infd+1, (SELECT_ARG_2_T) &fds,
                       NULL, NULL,&select_tv)) < 1) {
    if (errno != EINTR || errflag || retflag || breaks || contflag)
      break;
  }

Here select() keeps returning 0, indicating timeout. This is not an
error, so errno doesn't get set. If it was EINTR prior to the call,
it stays EINTR, and the loop keeps spinning.

As a workaround, powerlevel10k sets errno to ENOTTY (any value other
than EINTR will do) prior to calling sysread with timeout.

1 files changed, 2 insertions, 0 deletions

diff --git a/internal/worker.zsh b/internal/worker.zsh
index ce38af98..cca786fc 100644
--- a/internal/worker.zsh
+++ b/internal/worker.zsh
@@ -34,6 +34,7 @@ function _p9k_worker_main() {
         if [[ $fd == 0 ]]; then
           local buf=
           while true; do
+            [[ -t 0 ]]
             sysread -t 0 'buf[$#buf+1]'  && continue
             (( $? == 4 ))                || return
             [[ $buf[-1] == (|$'\x1e') ]] && break
@@ -115,6 +116,7 @@ function _p9k_worker_receive() {
 
     local buf resp
     while true; do
+      [[ -t $_p9k__worker_resp_fd ]]
       sysread -t 0 -i $_p9k__worker_resp_fd 'buf[$#buf+1]' && continue
       (( $? == 4 ))                                                                   || return
       [[ $buf == (|*$'\x1e')$'\x05'# ]] && break