Conformance and testing

Much of the behavior of webcolors is dictated by the relevant Web standards, which define the acceptable color formats, how to determine valid values for each format and the values corresponding to defined color names. Maintaining correct conversions and conformance to those standards is crucial.

The normal test suite

The normal test suite for webcolors – that is, the set of unit tests which will execute using standard Python test runners – aims for 100% code coverage, but does not aim for 100% coverage of possible color value inputs and outputs. Instead, it uses a small number of test values to routinely exercise various functions.

The test values used in most test functions are chosen to provide, where applicable, at least one of each of the following types of values:

  • An endpoint of the acceptable range of values (i.e., #ffffff and/or #000000 for hexadecimal).
  • A value beyond the high end of the acceptable range (i.e., greater than 255 in an integer triplet, or greater than 100% for a percentage triplet).
  • A value beyond the low end of the acceptable range (i.e., less than 0 in an integer triplet, or less than 0% for a percentage triplet).
  • A “negative zero” value (-0 in an integer triplet, or -0% in a percentage triplet).
  • An arbitrary value not from an endpoint of the acceptable range (usually #000080, chosen because the author likes navy blue).
  • A value which corresponds to a named color in CSS 3/SVG but not in earlier standards (usually #daa520, which is goldenrod in CSS 3/SVG).

Since this covers the cases most likely to produce problems, it provides good basic confidence in the correctness of the tested functions.

However, the normal test suite cannot guarantee that the color definitions included in webcolors correspond to those in the relevant standards, and cannot provide guarantees of correct conversions for all possible values. For that, additional tests are required.

Those tests are contained in two files in the source distribution, which are not executed during normal test runs: tests/definitions.py and tests/full_colors.py.

Verifying color definitions

The definitions test file verifies that the color definitions in webcolors are correct. It does this by retrieving the relevant standards documents as HTML, parsing out the color definitions in them, and comparing them to the definitions in webcolors. That consists of:

  • Parsing out the names and hexadecimal values of the 16 named colors in the HTML 4 standard, and checking that the names and values in HTML4_NAMES_TO_HEX match.
  • Parsing out the names and hexadecimal values of the 17 named colors in the CSS 2.1 standard, and checking that the names and values in CSS21_NAMES_TO_HEX match.
  • Parsing out the names and hexadecimal and integer values of the 147 named colors in the CSS 3 color module (although the color set is taken from SVG, CSS 3 provides both hexadecimal and integer values for them, while the SVG standard provides only integer values), and checking that the names and values in CSS3_NAMES_TO_HEX match, and that name_to_rgb() returns the correct integer values.

The definitions file can be run standalone (i.e., python tests/definitions.py) to execute these tests, but it does require an internet connection (to retrieve the standards documents) and requires the BeautifulSoup library for HTML parsing.

Fully verifying correctness of conversions

The full_colors test file exercises hex_to_rgb(), rgb_to_hex(), rgb_to_rgb_percent() and rgb_percent_to_rgb() as fully as is practical.

For conversions between hexadecimal and integer rgb(), that file generates all 16,777,216 possible color values for each format in order (starting at #000000 and (0, 0, 0) and incrementing), and verifies that each one converts to the corresponding value in the other format. Thus, it is possible to be confident that webcolors provides correct conversions between all possible color values in those formats.

Testing the correctness of conversion to and from percentage rgb(), however, is more difficult, and a full test is not provided, for two reasons:

  1. Because percentage rgb() values can make use of floating-point values, and because standard floating-point types in most common programming languages (Python included) are inherently imprecise, exact verification is not possible.
  2. The only rigorous definition of the format of a percentage value is in CSS 2, which declares a percentage to be “a <number> immediately followed by ‘%’”. The CSS 2 definition of a number places no limit on the length past the decimal point, and appears to be declaring any real number as a valid value. As the subset of reals in the range 0.0 to 100.0 is uncountably infinite, testing all legal values is not possible on current hardware in any reasonable amount of time.

Since precise correctness and completeness are not achievable, webcolors instead aims to achieve consistency in conversions. Specifically, the full_colors test generates all 16,777,216 integer rgb() triplets, and for each such triplet t verifies that the following assertion holds:

t == rgb_percent_to_rgb(rgb_to_rgb_percent(t))

The full_colors test has no external dependencies other than Python, and does not require an internet connection. It is written to be run standalone (python tests/full_colors.py). However, due to the fact that it must generate all 16,777,216 color values multiple times, and perform checks on each one, it does take some time to run even on fast hardware.