[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Scheme-reports] Requesting early review of a character sequence library
Jonathan A Rees scripsit:
> Can you say what your goals are?
Here's the content of the Rationale section:
When SRFI 13 was defined in 1999, it was intended to provide
efficient string operations on both whole strings and substrings. At
that time, only Guile and T provided true shared copy-on-write
substrings, and SRFI 13 could not reasonably require them of a Scheme
implementation. Consequently, almost all the SRFI 13 procedures accept
optional start and end arguments for each of the string arguments,
indexing the beginning and the end of the substring(s) to be operated on.
Unfortunately, variable-arity procedures are often slow and may not
interact well with type checking in Schemes that provide it. In addition,
it is now fairly common to store strings internally as UTF-8 or UTF-16
code unit sequences, which means that indexing operations are often O(n)
rather than O(1), and string mutation can be extremely expensive if the
storage used for the string needs to be expanded and the implementation
does not use an indirect pointer to it (as in Chicken).
As for shared substrings, they are no more common in 2015 than they were
in 1999. Fortunately, however, since then it has become normal for Schemes
to provide user-defined records, and they are required by both R6RS and
R7RS. This makes it possible to portably provide a representation for
a segment of a string, provided the string is never mutated. The most
portable such record consists of a string and two indexes, but other
more efficient representations may be used instead.
This proposal, therefore, is intended to help move the practice of Scheme
programming away from mutable strings, string indexes, and SRFI 13,
while maintaining as much backward compatibility as is consistent with
these goals. It does not require any particular run-time efficiencies
from its procedures. The string procedures, as well as string-transform,
make it possible to migrate a code base gradually.
It is also possible to implement character spans as ropes (trees of
strings), which makes concatenation more efficient at the expense of
more complex cursor objects and/or slower conversion to strings. For
this reason, as well as for security and efficiency reasons, there is
no operation to retrieve an underlying string from a character span,
as there may be more than one such string or none at all. The operations
provided here (with the exception of those in the Compatibility section)
are entirely independent of the character repertoire supported by the
implementation. In particular, this means that the case-insensitive
procedures of SRFI 13 are excluded. There is also no provision for R6RS
normalization procedures or for an string->integer procedure that was
proposed for SRFI 13 but not included. These may appear in future SRFIs.
John Cowan http://www.ccil.org/~cowan cowan@x
Go, and never darken my towels again!
--Rufus T. Firefly
Scheme-reports mailing list