storborg / manhattan Goto Github PK

It would be cool to be able to see show tests that have a result, and/or sort them by significance.

Should be more tolerant to an incomplete environment setup.

Currently an update and possibly an insert is issued for every new record persisted in the SQL store. There's probably a performance gain to be had from doing an INSERT ... ON DUPLICATE KEY UPDATE (or for sqlite, INSERT ... ON CONFLICT REPLACE).

There are some more tips about making it sqlalchemy-idiomatic here:
http://www.sqlalchemy.org/trac/ticket/960

The 'mh' cookie is used by RIM proxies, and causes problems with clients on the RIM tablet OS.

Should be more tolerant to an incomplete environment setup.

This probably means not using base64 for cookie encoding, but may also have other implications.

It shouldn't ever be necessary to access it in javascript.

When path is a bare file name (i.e., it has no directory prefix), TimeRotatingLog.create_dirs bails when trying to create a parent directory with no name.

Possible solutions:

• Convert path to absolute path in __init__
• Skip creating parent directories if dirpath is empty in create_dirs

Instead of having a default, middleware secret should be mandatory, so that it actually gets set to something different.

Now that the cookie handling quirks are more sorted out, it probably makes sense to make the whole system more robust to malformed cookies. If a cookie sig doesn't match, just give up and set a new cookie.

Keep a consistently-updated table containing just a certain number or certain time period of recent sessions. For icing on the cake, allow keeping specific tables for sessions that have recently converted on a certain goal.

Add backend method(s) to view the history for a given visitor id.

As of now the frontend and logs support site-specific data, but the backend aggregator does not.

Denormalized results should be available per-site, using the .site_id attribute available on each record. This could be implemented either:

• Explicitly in each Rollup instance as desired
• Implicitly by building additional rollups for each site
• Implicitly by performing each rollup for each site
• ???

Save on roundtrip times by adding client/server functionality to combine multiple backend queries into the same request.

The thread-based approach is obnoxious to test, because incorrectly behaving threads may live longer than expected, and synchronization is difficult.

Possible solutions could be:

• Switch the server/worker/log stack to gevent, so that Greenlets can be killed reliably.
• Use proper synchronization mechanisms, like threading Events / locks / semaphores for various steps of processing.
• Eliminate the need to spawn threads at all in tests.

Exclude bots by user agent regex matching and/or IP address / subnet matching.

In the middleware, the exception thrown on a signature mismatch when calling self.signer.unsign should include a message with both the desired signature and actual signature, instead of just the actual signature. This should aid in diagnosing issues with cookie mangling, truncation, etc.