git @ Cat's Eye Technologies Falderal / master doc / py-falderal.markdown

Tree @master (Download .tar.gz)

py-falderal.markdown @masterview markup · raw · history · blame


py-falderal is an implementation of Falderal in Python 2.5.x.


The original reference implementation of Falderal was written in Haskell. There are a few reasons I had for re-implementing it in Python:

  • The original Falderal implementation grew out of a Haskell-specific hack, and it showed in how it was written.

  • Fewer discrepancies between platforms. In particular, ghc for Windows likes to operate in terms of MS-DOS end-of-lines (CR, LF), but I tend to use it under Cygwin using Unix end-of-lines (LF).

  • Smaller install burden: Python sometimes comes bundled with the operating system; Haskell rarely does.

  • Haskell, being lazy, makes it harder to deal with exceptions; unless the Haskell expression is being evaluated both strictly and deeply, exceptions can slip by. Haskell's facilities for forcing an expression to be evaluated strictly and deeply are rather hacky (last I checked, the most portable way to do so was to use show.) For Falderal's purposes, this limitation seems artificial, at best.

  • Relatedly, Python (or CPython, anyway) has better error behavior than Haskell (or ghc, anyway); when it crashes, it dumps a backtrace (which I can then analyze), instead of just saying something pithy like Prelude: empty list (which I can't.)


py-falderal implements a slightly different subset of the Falderal Literate Test Format than the Haskell implementation did.

In particular,

  • It mainly implements shell command implementations. In practice, partly due to the "strict & deep" evaluation problem mentioned above, that's how I've been using Falderal anyway; also, its approach makes it somewhat more useful for "end-to-end" testing of compilers and interpreters, than for unit-testing individual text-processing functions inside a program. (Technically, it implements Haskell function implementations too, but it's something of a hack that uses ghc -e.)

  • I plan for it to only understand indented Falderal blocks. The idea is that the Falderal tests will almost certainly be embedded in a Markdown document (possibly with Bird-style literate code also embedded therein,) and no extra processing should be required to format something readable from that Markdown. The four-space-indented Falderal blocks come out as preformatted blocks, which is quite good enough. Relatedly,

  • It does no formatting. There are a set of classes for parsing Falderal files (Documents) into blocks (Blocks), and methods for formatting blocks into text. But aside from reporting the results of a test run, they're not used by the utility to format Falderal files. If you want to do extra processing on your Falderal/Markdown file so that Falderal blocks are prettier, you certainly can do that, but you'll need to write your own script which uses these classes, for it is outside the scope of py-falderal.

  • I'm not so sure about -> Functionality "blah" is implemented by shell command "flargh" anymore. I mean, it should certainly be the case that functionalities can have multiple implementations, but

    • Perhaps implementations should not be specified in Falderal documents at all — that approach is more abstract. But it requires them to be specified on the tool's command line, which in practice requires there to be a driver script to run the tests, for a particular implementation; but this is not necessarily a bad thing. (The effective practice has been, actually, to write out a bit of "configuration Markdown" with that "Functionality is implemented by" snippet in it, and to load that before loading the main tests. Rather than to configure the functionality implementation(s) on the command line itself.)
    • When falderal is run on more than one input file, what is the scope of a functionality, and what is the scope of the implementations of a functionality? Currently, in the Falderal semantics, that scope is global, and maybe that is appropriate; but maybe finer-grained control would be nice.
  • py-falderal also does not try to optimize the test runs into a block of test runs (which didn't work out so well in the Haskell implementation, for shell commands, anyway.)