Possible bug in Library domain for matrices

[Tobias Neumann]

This seems like a bug in Library to me:

Working example:

matx1 := matrix([[0,x1],[0,0]]) :: Matrix(Expression(Integer))

lib := library("mat.lib")

lib.'matx1 := matx1;

lib.'matx1

Failing to read back when somehow when there is a non-trivial kernel in the matrix, like sqrt(x), or exp(x), etc..

matx1 := matrix([[0,sqrt(x1)],[0,0]]) :: Matrix(Expression(Integer))

lib := library("mat.lib")

lib.'matx1 := matx1;

lib.'matx1

lib.'matx1

   >> System error:

   illegal sharp macro character: #\<

  Line: 139465, Column: 32, File-Position: 13113927

  Stream: #<SB-SYS:FD-STREAM for "file /home/tobias/mat.lib/index.KAF" {1003932393}>

[Ralf Hemmecke]

Hi Tobias,

I cannot help you with the Library domain, but perhaps the new Format1D
domain can help you. (See current github repo, you would have to compile
from the HEAD there.)

Format1D is not perfect, but I have created it in order to cut&paste
output from FriCAS back to the input line. I would be happy if someone
tries it out and sends feedback.

Switching it on is a bit cumbersome, but not too hard.

)set output formatted on
F ==> Formatter
l := [F Format1D];
setFormats!(l pretend List(OutputFormatterCategory))$FormattedOutput

If you do not like algebra output together with Format1D, then switch it
off.

)set output algebra off


(1) -> )set output formatted on
(1) -> F ==> Formatter
Type: Void
(2) -> l := [F Format1D];
Type: List(Type)
(3) -> setFormats!(l pretend List(OutputFormatterCategory))$FormattedOutput
Type: Void
(4) -> matx1 := matrix([[0,x1],[0,0]]) :: Matrix(Expression(Integer))

+0 x1+
(4) | |
+0 0 +
matrix[[0, x1], [0, 0]]

Type: Matrix(Expression(Integer))
(5) -> matx2 := matrix([[0,sqrt(x1)],[0,0]]) :: Matrix(Expression(Integer))

+ +--++
(5) |0 \|x1 |
| |
+0 0 +
matrix[[0, sqrt(x1)], [0, 0]]

Type: Matrix(Expression(Integer))
(6) -> matrix[[0, sqrt(x1)], [0, 0]]

+ +--++
(6) |0 \|x1 |
| |
+0 0 +
matrix[[0, sqrt(x1)], [0, 0]]

Type: Matrix(Expression(Integer))



You could also do a few tricks with it as is done in JFriCASSupport

(9) -> setFormat!(Format1D)$JFriCASSupport
Type: Void
(12) -> )set output algebra off
(12) -> matx1 := matrix([[0,x1],[0,0]]) :: Matrix(Expression(Integer))

--FORMAT:BEG:Format1D
matrix[[0, x1], [0, 0]]
--FORMAT:END:Format1D

Type: Matrix(Expression(Integer))
(13) -> matx2 := matrix([[0,sqrt(x1)],[0,0]]) :: Matrix(Expression(Integer))

--FORMAT:BEG:Format1D
matrix[[0, sqrt(x1)], [0, 0]]
--FORMAT:END:Format1D
Type: Matrix(Expression(Integer))

So you could parse the output easily by a script.
The strings between the --FORMAT:BEG ... --FORMAT:END should be readable
as if they come from a .input file.


Ralf

[Tobias Neumann]

Hi Ralf,

thanks, this is great. While this is not entirely what I am looking for in my specific case,

I was looking for such an output format before. (I read in a few MB large matrix expression

and wanted to check if saving and loading it from a Library would be faster).

In some cases I actually prefer an output format like Format1D over the default 2D.

As for the Library: I think it's useful anyhow to have such bugs collected/reported somewhere.

I've written a Mathematica interface to FriCAS a few days ago, and for that

I'm using unparse( stuff :: InputForm ) to return stuff to Mathematica. Your new

Format1D output format might actually be somewhat better for that (just as you do for jfricas).

Tobias

[Waldek Hebisch]

On Fri, Apr 23, 2021 at 12:02:55PM -0700, Tobias Neumann wrote:
> This seems like a bug in Library to me:
>

> Failing to read back when somehow when there is a non-trivial kernel in the
> matrix, like sqrt(x), or exp(x), etc..
>
> matx1 := matrix([[0,sqrt(x1)],[0,0]]) :: Matrix(Expression(Integer))
> lib := library("mat.lib")
> lib.'matx1 := matx1;
> lib.'matx1
>
> lib.'matx1
>
> >> System error:
> illegal sharp macro character: #\<
>
> Line: 139465, Column: 32, File-Position: 13113927
>
> Stream: #<SB-SYS:FD-STREAM for "file /home/tobias/mat.lib/index.KAF"
> {1003932393}>

Yes, there was rather general problem with arrays, needed code was
missing. Should be fixed now in the trunk.

--
Waldek Hebisch