- use mmap() to read whole files in core instead of allocating memory
and read'ing it.
- use a new, more general, HTML parser (html-parse.c) and interface to
it from Wget (html-url.c).
- respect <meta name=robots content=nofollow> (easy with the new HTML
parser).
- use hash tables instead of linked lists in places where the lists
were used to facilitate mappings.
- rewrite the code in host.c to be more readable and faster (hash
tables instead of home-grown lists.)
- make convert_links properly convert partial URLs to complete ones
for those URLs that have *not* been downloaded.
- use HTTP persistent connections where available. very
simple-minded, caches the last connection to the server.
Published in <sxshf533d5r.fsf@florida.arsdigita.de>.
URLs, gen_page.cgi?page1 and get_page.cgi?page2, they'll both be saved as
get_page.cgi and the second will overwrite the first. Also, parameters to
implicit CGIs, like "http://www.host.com/db/?2000-03-02" cause the URLs to be
printed with trailing garbage characters, and could seg fault. I'm not sure
what Dan had in mind with this patch (no explanatory comments), but I'm removing
it for now. If he can rewrite it so it doesn't break stuff, okay.
Got rid of newly-introduced nested-if warnings in ftp.c and http.c. Fixed
apparently completely untested code in main.c that was trying to provide --wait
/ --waitretry backwards compatibility, but had multiple fundamental bugs.
together, we compare local file X.orig (if extant) against server file X.
Previously -k and -N were worthless in combination because the local converted
files always differed from the server versions.
