On Mon, Jun 20, 2016 at 10:14:52PM -0400, Michael Silver wrote:
Alternatively, we could decide to do this for CHICKEN 5 only, and rename
both procedures to match the Guile naming convention. Maybe do a survey
of existing implementations that offer such a procedure and see what names
they use.
Maybe, but even coming from someone who is porting a huge Guile codebase to CHICKEN, I don’t mind CHICKEN having a different naming for the function. Find and replace is easy :)
Scheme xref says with-error-to-port is supported by Gauche, SCM and Stk:
http://practical-scheme.net/wiliki/schemexref.cgi?with-error-to-portversus ours, which is only in CHICKEN:
http://practical-scheme.net/wiliki/schemexref.cgi?with-error-output-to-port
I don't know how complete the xref is. Obviously it's not comprehensive,
because it doesn't mention Guile on the page about with-error-to-string.
But it seems to have the most important implementations.
But it's all rather arbitrary: why is there no with-error-to-file?
And if there were, it would be jarring to have to (import scheme) to
get with-output-to-file, but have to do (import (chicken ports)) to get
with-error-to-file…
Makes sense, but since with-output-to-string requires importing ports, I don’t think its unreasonable to ask for the same for it’s sister function with-error-output-to-string. It makes as much sense to group those two functions together in one unit as it does to group with-output-to-port and with-error-output-to-port, which seems totally fine to me. The ports unit doesn’t seem to have anything relating to file output, but it does have both output and error output to ports, and output to string, but misses error output to string. Looked like an obvious “what’s missing from the pattern” to me, which is why I suggested the patch.
Yeah, it makes the most sense to at least make our own extensions to
the standard internally consistent, even if the standard itself is not
complete in this aspect (r5rs doesn't have current-error-port, r7rs does
have current-error-port but, oddly, no procedures to manipulate it).
Maybe we should add with-error-to-file to chicken.base, or whereever we
put current-error-port?
The attached patch adds with-error-to-string and renames
with-error-output-to-port to with-error-to-port, for chicken-5 only.
We probably shouldn't be adding new procedures to CHICKEN 4 at this
point: any CHICKEN 4 egg that used this would be depending on exactly
one minor release of CHICKEN that supports this.
The patch also tweaks types.db a bit to improve the type for the
*-to-string procedures, indicating they always return one value,
which is always a string.
Thanks for your suggestion, Michael!
Cheers,
Peter
<0001-Add-with-error-to-string-and-rename-with-error-outpu.patch>
smime.p7s