branch: externals/plz
commit 0a860d94dcbb103d05f3ee006772a568904fa4de
Author: Adam Porter <a...@alphapapa.net>
Commit: Adam Porter <a...@alphapapa.net>
Tests: "There be dragons."
---
tests/test-plz.el | 62 ++++++++++++++++++++++++++++++++++++++++++++++++++++++-
1 file changed, 61 insertions(+), 1 deletion(-)
diff --git a/tests/test-plz.el b/tests/test-plz.el
index 87294d1af8..e219a142e9 100644
--- a/tests/test-plz.el
+++ b/tests/test-plz.el
@@ -20,7 +20,67 @@
;;; Commentary:
-;;
+;; NOTE: NOTE: NOTE: NOTE: Yes, four NOTEs, because this is important:
+;; As of this moment, all of the tests pass when run with makem.sh.
+;; And when running them in an interactive Emacs with ERT, one test at
+;; a time, individual tests pass, or almost always do (depending on
+;; whether the httpbin.org server is overloaded). But when running
+;; multiple tests in ERT at one time,
+;; i.e. (ert-run-tests-interactively "plz-"), multiple, if not most,
+;; tests fail, but not the same ones every time.
+
+;; I have now spent hours trying to figure out why, inserting many
+;; debug statements in many functions, and come up with nothing. I
+;; tried changing the way `accept-process-output' is called, like
+;; using timeouts or JUST-THIS-ONE, but it made no difference. I
+;; tried calling it extra times, nope. I tried calling the sentinel
+;; extra times when it seemed that it hadn't run the THEN function,
+;; nope. Nothing seems to make a difference.
+
+;; I even checked out an earlier commit, before the commit that
+;; rewrote/merged the synchronous request code into the `plz'
+;; function, thinking that surely I broke something--but, nope, they
+;; apparently failed the same way back then: passing with makem.sh,
+;; passing individually, but failing when run in close succession by
+;; ERT.
+
+;; After inserting enough debug statements, I noticed that the process
+;; sentinel sometimes seemed to run for the last time after the ERT
+;; test had returned, which suggests that ERT might be doing something
+;; weird, or somehow its instrumentation interferes with the
+;; process-handling code. But if that's not the cause, then I'm out
+;; of ideas.
+
+;; So then I tried rewriting the synchronous request code to use
+;; `call-process-region', instead of calling `accept-process-output'
+;; in a loop to block on the curl process (which is how the Elisp
+;; manual says to do it), but that still made no difference: even the
+;; async requests fail in the same way with ERT. So that doesn't
+;; appear to be the problem, either.
+
+;; So is there some kind of fundamental flaw in the `plz' design?
+;; Maybe. Is there a simple, logical oversight in its code that only
+;; manifests under certain conditions? Maybe. Is ERT doing something
+;; weird that's interfering with process-related code? Maybe. Is
+;; Emacs's own process-handling code still broken in some mysterious
+;; way? Maybe.
+
+;; But despite all of that, when using `plz' "in anger", in `ement',
+;; it seems to work reliably for me. I did get one report from one
+;; user that sounded like the same kind of problem I'm seeing with ERT
+;; here, but then he tried `ement-connect' again, and it worked. And
+;; I'm sitting here watching `ement' constantly using `plz' to talk to
+;; the matrix.org server, and I haven't had a single error or failure,
+;; even after hours of being connected. It *seems* to *actually*
+;; work.
+
+;; So, if you're reading this, and you're wondering whether you should
+;; use `plz': Well, please do, and please let me know if you have any
+;; problems; I do need to know whether it's working for other users.
+;; And if you think you might know what's going wrong when running the
+;; tests in ERT, please let me know, because I'm out of ideas: as far
+;; as I can tell, when it comes to process-handling in Emacs, "there
+;; be dragons."
;;; Code:
