Anonymous user
Talk:S-expressions: Difference between revisions
→practical solutions preferred
(→J) |
|||
Line 12:
you should ask yourself: would i use this approach in a real project? if a solution that you would really use is different from the task, please post it anyways and explain why.--[[User:EMBee|eMBee]] 04:01, 19 October 2011 (UTC)
::I used S-exps in a C++ project about eight years ago. More or less real S-exps with interned symbols, cons cells, etc. No garbage collection was used, but rather smart pointers with reference counting. This was used for inter-process communication over sockets, as well as for storing and retrieving complex configurations easily. The project was a mail transport. Later when an IMAP4 interface had to be developed, one of the other programmers realized that IMAP4 uses S-exps. We easily adapted the library to parse and generate IMAP4 which simplified the handling of that protocol greatly. Its commands and responses could be handled on a higher level as objects, using Lisp-like functions (in C++). None of us had known that IMAP4 uses S-exps, so we were amazed and the programmer coding the IMAP4 interface was pretty happy. (Some conventions had to be accomodated in our reader: IMAP4 has backslashes on symbols like \SEEN (flag indicating read). This is not an escape sequence but a constituent of the symbol name.) I had a package concept implemented in symbols: symbols belonged to packages (namespaces). This was became really useful for the IMAP4 module, because it was able to read forms in a private package to prevent symbols from polluting the regular namespace. Later in the project, we were given a requirement to develop some test tools that could be run on the mobile device to do some automated tests. I developed an evaluator over the S-expressions, creating a multi-threaded Lisp dialect with dynamic closures. I added a GUI where you could type expressions right on the device and hit EVAL. This saved them time because they didn't have to re-flash the whole image just to fix a bug in a script: upload the script and (load ...). Only if they needed a new intrinsic function in the interpreter did they have to rebuild the image and re-flash. Developing this dialect and the GUI window, and a concise, clear reference manual, took about three days (having the Lispy library in place). QA people developed multi-threaded tests which exercised various platform functions on the device (via intrinsic function bindings exposed in the dialect). Nobody in QA came to ask any major questions about the language, even though they had no Lisp experience, since the reference manual was thorough, and explained everything. What is a form, what is a variable, what is a binding, what is an environment, what is scoping, what does it mean to evaluate, what is a function, what is a closure, etc. '''Would I use S-exps as defined by this task?''' Not likely. You really need the correspondence to a proper Lisp-like structure which distinguishes symbols (interned objects) from strings. Interned symbols allow for identification using simple equality <code>if (car(form) == foo_s) ... // if the form begins with the foo symbol ... </code>. Strings and symbols really have to be a distinct data type, otherwise you're wasting cycles on string processing, and you have hygiene issues. If you need a unique symbol, just construct a symbol object and you got it (since it has a unique address). If symbols are strings, then you cannot make a unique symbol, only a unique-ish one based on giving the string contents that are unlikely to coincide with any other symbol. A lot of the power came from the simplicity of the API, not from what the notation looks like. So whether I would use a given S-exps solution in a project would depend heavily on the quality of the API used to manipulate the data structure corresponding to the S-exp. I would not use a solution that required me to manage memory with malloc and free, for instance, or which had some awfully encapsulated data structures with clumsy accessors. In this regard, some particular '''solutions''' to this task may be more useful than others. I would definitely not use the C code for instance. It is just an example. For a real task, you need a well-designed, ideally mature library for this type of thing, with some kind of garbage collection. [[Special:Contributions/192.139.122.42|192.139.122.42]] 23:06, 19 October 2011 (UTC)
::''I cannot imagine any case where I would use this for practical purposes. I already have more robust serialization and deserialization available with language primitives,''
| |||