| 1 | EXIM'S BEHAVIOUR CHANGES WHEN RUNNING IN THE TEST HARNESS |
| 2 | --------------------------------------------------------- |
| 3 | |
| 4 | When Exim is running in its test harness, via the scripts in the exim-testsuite |
| 5 | distribution, its behaviour is altered in a few ways, in order to make the |
| 6 | regression testing work properly. The documentation for the test suite |
| 7 | describes how a copy of the Exim binary is taken and patched in order to get it |
| 8 | to run in the test harness. This document briefly lists the behavioural changes |
| 9 | that result. They come into play when the Boolean variable running_in_test_ |
| 10 | harness is true. |
| 11 | |
| 12 | |
| 13 | Privilege |
| 14 | --------- |
| 15 | |
| 16 | Exim does not give up its root privilege when called with -C or -D, nor does it |
| 17 | insist on the caller being an admin user when starting a daemon, a queue |
| 18 | runner, or requesting debug output. |
| 19 | |
| 20 | |
| 21 | Small Pauses |
| 22 | ------------ |
| 23 | |
| 24 | In a number of places, typically when a subprocess has been forked, there are |
| 25 | short pauses of half or one second in one of the processes. This allows the |
| 26 | other process to "go first"; it ensures that debugging or logging output always |
| 27 | appears in the same order. |
| 28 | |
| 29 | |
| 30 | Daemon |
| 31 | ------ |
| 32 | |
| 33 | The daemon always writes a pid file when running in the test harness. |
| 34 | |
| 35 | |
| 36 | CRAM-MD5 |
| 37 | -------- |
| 38 | |
| 39 | The cram_md5 authenticator always uses the same challenge string. |
| 40 | |
| 41 | |
| 42 | Appendfile |
| 43 | ---------- |
| 44 | |
| 45 | After a quota error, the "time since last read" for the file is forced to 10s, |
| 46 | for repeatability. |
| 47 | |
| 48 | |
| 49 | Memory management |
| 50 | ----------------- |
| 51 | |
| 52 | Memory management debugging output contains only the store pool and the size |
| 53 | (other information is too variable). New memory is initialized to contain F0 in |
| 54 | all bytes. |
| 55 | |
| 56 | |
| 57 | Queue running |
| 58 | ------------- |
| 59 | |
| 60 | There's a facility (-Tqt) for fudging queue times for testing retry logic. |
| 61 | |
| 62 | |
| 63 | Syslog |
| 64 | ------ |
| 65 | |
| 66 | Exim never writes to syslog in the test harness. Attempts to do so are silently |
| 67 | ignored. None of the tests actually specify syslog logging for any actual log |
| 68 | lines, but there is one that tests the inability to open the main and panic |
| 69 | logs, which by default then tries to write to syslog. |
| 70 | |
| 71 | |
| 72 | SMTP connection timeout |
| 73 | ----------------------- |
| 74 | |
| 75 | In order to be able to test timeout handling, a "connection refused" error is |
| 76 | converted into a timeout if the timeout value is set to 999999s. |
| 77 | |
| 78 | |
| 79 | Random numbers |
| 80 | -------------- |
| 81 | |
| 82 | The seed for the pseudo-random number generator is set to a fixed value in the |
| 83 | test harness, to ensure repeatability. |
| 84 | |
| 85 | |
| 86 | Bounce messages |
| 87 | --------------- |
| 88 | |
| 89 | When Exim is submitting a bounce message to itself, unless the configuration |
| 90 | has set queue_only, it uses -odi so that the bounce is delivered before the |
| 91 | subprocess returns. This avoids a race that might put log lines in an arbitrary |
| 92 | order. |
| 93 | |
| 94 | |
| 95 | DNS lookups |
| 96 | ----------- |
| 97 | |
| 98 | The real DNS resolver is never called. Instead, a fake resolver, which runs as |
| 99 | a separate program, is used. It is part of the test suite and is documented |
| 100 | there. This ensures complete control over the exact results of any DNS lookups. |
| 101 | |
| 102 | An attempt to look up a PTR record for 99.99.99.99 or an IP address for a host |
| 103 | whose name ends with .test.again.dns always yields a "try again" error. |
| 104 | |
| 105 | A fake function is called instead of gethostbyname(). It recognizes the name |
| 106 | "manyhome.test.ex" and generates a humungous number of IP addresses. It also |
| 107 | recognizes an unqualified "localhost" and forces it to the appropriate loopback |
| 108 | address (IPv4 or IPv6, as required). IP addresses are treated as literals. For |
| 109 | other names, it does a DNS lookup (which of course actually calls the fake |
| 110 | resolver) to find the host name. |
| 111 | |
| 112 | |
| 113 | User names |
| 114 | ---------- |
| 115 | |
| 116 | If unknown_login is set, it forces the login name, thus overriding the actual |
| 117 | login for the test suite caller. When this happens, unknown_username provides a |
| 118 | user name if it is set; otherwise an empty string is used. |
| 119 | |
| 120 | |
| 121 | Ident |
| 122 | ----- |
| 123 | |
| 124 | If -bh is used and both the sending host port and the incoming interface port |
| 125 | are supplied, an ident (RFC 1413) call is made for testing purposes. |
| 126 | |
| 127 | |
| 128 | Debug output |
| 129 | ------------ |
| 130 | |
| 131 | Debugging output from the function that waits for the clock to tick at an |
| 132 | appropriate resolution (before completing the arrival of a message, for |
| 133 | example) is suppressed because the fractions of seconds that it contains will |
| 134 | never be repeatable. |
| 135 | |
| 136 | |
| 137 | Philip Hazel |
| 138 | 15 February 2006 |