libera/#commonlisp - IRC Chatlog
Search
3:39:52
kakuhen
I have stumbled on yet another CCL bootstrapping problem, and it's way beyond my knowledge and understanding
3:40:31
kakuhen
while I was able to successfully bootstrap 1.12.1 from 1.12 on FreeBSD/amd64, Linux is giving me problems (with the same computer) when I attempt loading the bootstrap image
3:41:04
kakuhen
I'm getting dropped into the lisp kernel debugger in the middle of loading some streams fasl -- even though I received no warnings or errors when compiling ccl or the level-0 files
3:52:54
kakuhen
the documentation notes that it should be (in theory) possible to compile ccl from another cl implementation, but such a process is experimental and not officially supported
3:53:36
kakuhen
in any case, i should probably just be happy with the binaries i'm already provided -- the bootstrap attempt was more of an exercise to see if I could do it on linux, since the distribution I'm on does not offer any official packages for CCL, but it does for ECL
3:55:50
beach
kakuhen: rme told me that, while it would be possible to make it so that CCL can bootstrap from other Common Lisp implementations, "it is probably not worth the effort to make it happen" [or something to that effect].
4:01:06
beach
I think some day I need to turn the SICL bootstrapping procedure into an implementation-independent library, so that it can be used by other implementations too. But I currently have absolutely no idea how to do that.
6:59:39
pjb
beach: Perhaps doing something metalinguistically like in scheme? https://www.youtube.com/watch?v=SrKj4hYic5A
7:02:21
beach
Looks interesting. I'll watch it later. I am not sure how relevant it is to Common Lisp bootstrapping, though. I guess I'll find out. I mean, SICL bootstrapping is already pretty "meta".
7:04:46
beach
The first few phases of SICL bootstrapping can be seen as building a circular graph of classes (MOP classes and other essential classes) and generic functions (again MOP and other essential functions). And we do that by using the standard tools like DEFCLASS, DEFMETHOD, DEFCLASS, DEFUN, DEFMACRO, etc.
7:05:11
pjb
I imagine that a first step to bootstrap an implementation, is being able to implement and run it inside another implementation.
7:05:59
pjb
So in your current system the standard tools are the meta level for the new (bootstrapping) level.
7:07:38
beach
I don't see that as levels, no. Each phase is merely meant to make the object representation different. First, we use host objects, then bridge objects, and finally ersatz objects.
7:08:11
beach
There are no "levels" in the sense of limited functionality or expressiveness in any level.
7:08:23
pjb
But perhaps having those bootstrapping phases is only an artifact of the development process. I mean, what would prevent to use the current implementation to generate everything of the new implementation completely independently of the current environment, in one single phase?
7:10:13
beach
So the problem I solve is the following: If I do something like "make-instance" in the host, it will create an opaque host object. But I need to do that in order to create some SICL classes like T, standard-object, etc.
7:11:35
beach
I then make a new MAKE-INSTANCE etc. that creates instances of those classes that have slots that SICL needs.
7:12:19
beach
Then I define a new MAKE-INSTANCE that take those funny non-host-classshes and create more SICL-like objects called ersatz objects.
7:12:26
pjb
But instead, you could just generate a binary that will have a single make-instance in the new image.
7:13:15
pjb
Yes. I understand the step-by-step approach you use, but as a development/learning/testing process.
7:14:22
beach
The steps are not important as you point out, but I need several different first-class global environments, each one containing objects represented in a different way. And I need three different such representations.
7:15:44
beach
So I naturally divided the bootstrapping procedure into three (plus a few more for initial and final stuff) phases, each phase creating objects with a different representation in a different first-class global environment.
10:05:21
contrapunctus
Devon: someone also suggested Delta, once - https://github.com/dandavison/delta
10:12:14
Devon
contrapunctus: Delta is merely "A viewer for git and diff output" like emacs diff-mode.
10:17:57
Devon
contrapunctus: Emacs ediff is merely "a comprehensive visual interface to diff & patch" so neither of them actually differs from unix diff.
10:55:11
splittist
Devon: smarter by having knowledge of the underlying semantics of the string? Absolutely. The trick would be to make a tractable way to specify the various edit operations and their costs. I do sometimes wonder if speed is all that important for actual human-readable texts.
11:47:12
jmercouris
I would have thought that any stream other than t would result in it not printing
11:48:37
jmercouris
also with regards to the previous discussion, there is a diff tool that works with SEXP
11:52:43
polygon-op
I suppose that it could be written differently, i.e (defun write (object &key (stream-p nil stream) …) ) and act based on the fact whether stream-p is true
11:55:43
scymtym
what you expected (write … :string nil) to do is basically the function WRITE-TO-STRING
11:59:06
scymtym
i'm confused. shouldn't the comparison be with (with-output-to-string (stream) (write 'tomato :stream stream))? i thought that was the intended behavior
11:59:41
polygon-op
it should, that's just lack of understanding the difference between the return value and the stream output ,) /me carries with his day
12:01:34
scymtym
but what is the desired behavior if it is neither returning a string nor writing to a stream?
12:32:21
polygon-op
ACTION http://turtleware.eu/static/paste/0de83150-triangles.webm - "tracing" the triangulation algorithm in CLIM