[SECP256K1] Fix issue where travis does not show the logs

Summary:
```
I noticed that according to the doc tests.log should contain stdout as
well as stderr. But it doesn't because stdout isn't flushed. I removed
buffering completely to avoid having to call fflush twice.
```

Backport of secp256k1 [[bitcoin-core/secp256k1#685 | PR685]].

Depends on D5433.

It slightly varies from the PR by only printing the logs in the event of
a failure.

There is no need to explicitly cat the logs with the cmake build because
they are already printed by the test runner in case of failure.

Before: https://travis-ci.org/Fabcien/secp256k1/jobs/658610743
After: https://travis-ci.org/Fabcien/secp256k1/jobs/658631083

Test Plan:
Run the Travis build (see
https://travis-ci.org/Fabcien/secp256k1/builds/658633612).

Reviewers: #bitcoin_abc, deadalnix

Reviewed By: #bitcoin_abc, deadalnix

Subscribers: deadalnix

Differential Revision: https://reviews.bitcoinabc.org/D5434

Author: jonasnick

src/tests.c

@@ -5176,6 +5176,15 @@ void run_ecdsa_openssl(void) {
 int main(int argc, char **argv) {
     unsigned char seed16[16] = {0};
     unsigned char run32[32] = {0};
+
+    /* Disable buffering for stdout to improve reliability of getting
+     * diagnostic information. Happens right at the start of main because
+     * setbuf must be used before any other operation on the stream. */
+    setbuf(stdout, NULL);
+    /* Also disable buffering for stderr because it's not guaranteed that it's
+     * unbuffered on all systems. */
+    setbuf(stderr, NULL);
+
     /* find iteration count */
     if (argc > 1) {
         count = strtol(argv[1], NULL, 0);

travis/build_autotools.sh

@@ -35,6 +35,12 @@ pushd buildautotools
   $AUTOTOOLS_EXTRA_FLAGS \
   $USE_HOST
 
+print_logs() {
+  cat tests.log || :
+  cat exhaustive_tests.log || :
+}
+trap 'print_logs' ERR
+
 make -j2 $AUTOTOOLS_TARGET
 
 popd