random corpus generator, v3

[Ethan Glasser-Camp]

david at tethera.net writes:

> This obsoletes the series at:
>
>      id:"1344888831-4301-1-git-send-email-bremner at debian.org"
>
> Changes since v2:
>
> - clean up new test-binaries and objects
>
> - remove the "set -o pipefail" leftover from debugging.  Possibly this
>   makes sense as a global setting, but in a seperate patch.
>
> - add hex-escape to test/basic
>
> - rebase against updated master.

Hi! This looks pretty good to me and I am for improving the test
infrastructure.

Some minor problems:

- Patch 2 doesn't apply; neither do patches 4 or 5, presumably due to changes
  that weren't made due to patch 2.

- Commit message discipline: the subject line of patch 4 ends in a
  period. "Seperate" is spelled by most people as "separate", though I
  would encourage you to buck the trend if you are so inclined.

- In patch 4:

> +    if (private_status == NOTMUCH_PRIVATE_STATUS_NO_DOCUMENT_FOUND) {
> +       _notmuch_message_add_term (message, "type", "mail");
> +    } else {
> +       return NOTMUCH_STATUS_DUPLICATE_MESSAGE_ID;
> +    }

Why not switch the branches? That is, check for private_status !=
NOTMUCH_PRIVATE_STATUS_NO_DOCUMENT_FOUND and return immediately?

- In patch 5:

> +    for (count = 0; count < num_messages; count++) {
> +	int j;
> +	int num_tags = random () % (max_tags + 1);
> +	int this_mid_len = random () % message_id_len + 1;

This looks odd. I'm pretty sure it's correct, but my brain keeps saying,
"Why are there no parentheses on (message_id_len + 1)?" Maybe just a
comment that message ids must be at least one character long, or the
ranges of values necessary for both of these variables.

Ethan