text-convert

Convert between various types representing textual data. This library differs from the similar string-conversions library in that this library exports class methods that are monomorphic in their return type. This enhances readability and aids type inference.

The module Text.Convert exports seven one-method type classes. The methods are designed to be input-type polymorphic and output-type monomorphic, to aid with readability and type inference. The classes are:

• ToString, with method asString;
• ToByteString, with method asByteString;
• ToLazyByteString, with method asLazyByteString;
• ToByteStringBuilder, with method asByteStringBuilder;
• ToText, with method asText;
• ToLazyText, with method asLazyText; and
• ToTextBuilder, with method asTextBuilder.

Design goals in order of importance:

1. correctness (including totality);
2. efficiency; and
3. ease of use. The author hopes that the design achieves these goals to a high degree.

To comment on correctness, we are using lenient UTF-8 decoding when converting between ByteStrings and Texts, which means the conversions are correct in the sense that they are total, but they are incorrect in the sense that they may replace some unicode characters with the usual replacement character. Since this library serves a very general purpose, users will not necessarily place any restrictions on the inputs of these conversion functions, so the author thought that totality was important. Notably, this choice agrees with the behavior of many text editors and readers that use similar decoding strategies.

To comment on efficiency, we avoid unnecessary intermediate conversions where possible. For example, instead of converting Text to ByteString through String (a design admirable in that it is simple, correct, and a quantity of code that is linear in number of supported datatypes), we choose to use a bespoke conversion function for every pair of supported datatypes wherever possible (resulting in a quadratic quantity of code). We also avoid unnecessary allocation. For example, instead of converting LazyByteString to Text through LazyText (an approach that would entail creating a new rope of Text that would immediately be converted into a single array), we convert LazyByteString to Text through ByteString (which is a single array). We don't have any benchmarks to support this design choice, admittedly, as we consider the cost of creating a benchmark suite overrides the benefits of having one at this time. Should interest in this library grow, we may revisit this decision.