gdb/testsuite: remove guile "test byte at sp, before flush" test

When I run:

    $ while make check TESTS="gdb.guile/scm-ports.exp" RUNTESTFLAGS="--target_board=native-extended-gdbserver"; do done

I eventually get:

    FAIL: gdb.guile/scm-ports.exp: buffered=1: test byte at sp, before flush

I don't know why I only see this (or see this more often) with the
native-extended-gdbserver board, I guess because it changes the timing
of things.  Also, this is with a debug/ASan/UBSan build of GDB, if that
matters.  This is with guile 3.0.

This is what the test attempts to do:

 1. create a rw memory port that is supposed to be buffered
 2. write the byte at $sp with a new value, expecting that byte to only
    be in the port's buffer
 3. read the memory at $sp (using parse-and-eval, bypassing the rw
    memory port), expecting to see the old byte value
 4. flush the memory port, expecting that to write to new byte value at
    $sp
 5. read the memory at $sp (using parse-and-eval again), expecting to
    see the new value

It is step 3 that fails.  My hypothesis is that even if the memory port
we write to is buffered, the write to the underlying memory can happen
before the "before flush" test happens.  The Guile Buffering doc [1]
says this:

    If you write data to a buffered port, it probably doesn’t go out to
    the mutable store directly. (This “probably” introduces some
    indeterminism in your program: what goes to the store, and when,
    depends on how full the buffer is. It is something that the user
    needs to explicitly be aware of.) The data is written to the store
    later – when the buffer fills up due to another write, or when
    force-output is called, or when close-port is called, or when the
    program exits, or even when the garbage collector runs.

Because of the mention of the garbage collector, I tried putting some
calls to gc-disable and gc-enable around that area, but it doesn't fix
it.

Given that the flushing behavior of the buffered port seems
non-deterministic, I think that the buffering behavior can't easily be
tested.  I propose to get rid of that specific aspect of the test.

As suggested by Tom de Vries, replace the "test byte at sp, before
flush" test with a "test byte at sp, before write" one.  I think this is
useful as a sanity check to help prove that the memory write did in fact
change the state of the program.

Change-Id: I2f0afd7b2ebb7738e675f58397677f2f9a4e06bb
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29825
Reviewed-By: Thiago Jung Bauermann <thiago.bauermann@linaro.org>

diff --git a/gdb/testsuite/gdb.guile/scm-ports.exp b/gdb/testsuite/gdb.guile/scm-ports.exp
index 3cc8cde0af1ad8401b53829d453f1e88e22e66a4..484f7bcddd03e3a0d0bed52a09011a992d0c191d 100644 (file)
--- a/gdb/testsuite/gdb.guile/scm-ports.exp
+++ b/gdb/testsuite/gdb.guile/scm-ports.exp
@@ -113,20 +113,27 @@ proc test_mem_port_rw { buffered } {
        "define old-value"
     gdb_test_no_output "guile (define new-value (logxor old-value 1))" \
        "define new-value"
+
+    # Sanity check before the write.
+    gdb_test "guile (print (value=? (parse-and-eval \"*(char*) \$sp\") byte-at-sp))" \
+       "= #t" \
+       "test byte at sp, before write"
+
+    # Write new byte at *SP.
     gdb_test "guile (print (put-bytevector rw-mem-port (make-bytevector 1 new-value)))" \
        "= #<unspecified>"
+
+    # Force flush, necessary if the port is buffered.
     if {$buffered} {
-       # Value shouldn't be in memory yet.
-       gdb_test "guile (print (value=? (parse-and-eval \"*(char*) \$sp\") byte-at-sp))" \
-           "= #t" \
-           "test byte at sp, before flush"
        gdb_test_no_output "guile (force-output rw-mem-port)" \
            "flush port"
     }
+
     # Value should be in memory now.
     gdb_test "guile (print (value=? (parse-and-eval \"*(char*) \$sp\") byte-at-sp))" \
        "= #f" \
-       "test byte at sp, after flush"
+       "test byte at sp, after write"
+
     # Restore the value for cleanliness sake, and to verify close-port
     # flushes the buffer.
     gdb_test "guile (print (seek rw-mem-port (value->integer sp-reg) SEEK_SET))" \