lynx-dev lynx2.8.3dev.17

[T.E.Dickey]

I've a handful of configuration/porting issues, but would like to make the
next patch a pre-release.  That would incorporate only bug-fixes...

1999-12-15 (2.8.3dev.17)
* add a caution to jumpsUnix.html -PW
* ifdef LYEditKeyForAction() to compile when EXP_ALT_BINDINGS is not defined
  (reported by VH) -TD
* add 3-valued DEFAULT_KEYPAD_MODE to lynx.cfg -TD
* revert DEFAULT_KEYPAD_MODE to NUMBERS_AS_ARROWS (consensus) -TD
* add COOKIE_SAVE_FILE lynx.cfg and corresponding command-line option, default
  it to /dev/null, which is recognized on all systems -BJP
* correct typo in declaration of LYLeakSAVsprintf() -Michael Warner
* remove from_source_cache_p parameters in LYMainLoop.c (request by LP) -TD
* documentation updates covering the Options Menu & other changes omitted
  previously -PW
* small correction in TRSTable.c:  One of my previous changes for TRST broke
  inheritance of alignment for cell contents from TR elements -KW
* new functions for keeping track of temp files (and some other files) for user
  interface pages:  LYRegisterUIPage(), LYIsUIPage().  Most checks that look at
  the loaded document's title have been replaced by this mechanism.  This also
  replaces various mechanism that were implemented for some specific pages -KW
* also replaced some more checks that looked at the document's title by checks
  of the address, for pages that are always identified by a special URL scheme
  (LYNXCOOKIE:, LYNXKEYMAP:, LYNXMESSAGES:) -KW
* now check directly in postoptions() whether the loaded document is one from
  which submission of option changes can be accepted, using the new tracking
  mechanism.  Only the form-based Options Menu and Visited Links are allowed. 
  Relying on a good random generator to prevent spoofing is not necessary. 
  Kept the secure_string check, but it should be only regarded as an aid to the
  user, it prevents changing options from an instance of the form-based Options
  Menu that is not the most recent one (this can happen if an Options Menu page
  is pushed on the history stack and later returned to) -KW
* honor REUSE_TEMPFILES for some more user interface pages.
* prevent an obscure LYNXOPTIONS://MBM_MENU crash.
* other minor additions and corrections to checks for special pages -KW
* changed INSTALLATION text for --enable-internal-links -KW, HN
* changes to build dev16 with color styles on DJGPP/PDCurses 2.3
  (you must uncomment a line in makefiles to build lynx that way) -LP
* correct a missing check in HTML.c HTML_A case against force_empty_hrefless_a
  (reported by KW) -VH
* changed lines -> disp_lines, cols -> disp_cols in GridText.c -KW
* don't reparse with SOURCE_CACHE if screen size changed only
  vertically (disp_lines).  Also don't reparse unnecessarily in
  some cases of LYK_MINIMAL -KW
* don't append (HEAD) to title for INFO screen if already present -KW
* set newdoc.title to curdoc.title in LYMainLoop.c for SOURCE command.
  Otherwise INFO could show the wrong document's title -KW
* made global from_source_cache PRIVATE to LYMainLoop.c -KW
* reset DIRED_MENU item list when lynx.cfg is reloaded -KW
* new lynx.cfg option AUTO_UNCACHE_DIRLISTS for Dired mode, see lynx.cfg -KW
* prevent filedescriptor and inode leak in HTFWriter.c that occurred if
  the same resource is repeatedly downloaded (for 'd' or for passing to
  a viewer) -KW
* corrected logic for suppressing last HTDisplayPartial call at end of file,
  after discussion with LP -KW
* corrected line counting in HTDisplayPartial, taking into account offsets used
  for the various variables being compared -KW
* added confirmation prompt for reloading non-safe post_data documents in
  form-based Options Menu change handling.  Before this change, the prompt
  would be issued after returning to mainloop() instead, with different
  behavior if confirmation was denied:  The document would be popped instead of
  keeping the current instance -KW
* function confirm_post_resub made PUBLIC and moved to HTAlert.c.  New function
  srcmode_for_next_retrieval in LYGetFile.c.  Both used for the above and in
  LYMainLoop.c -KW
* fixed behavior for various post_data confirmation prompts in connection with
  SOURCE_CACHE in LYMainLoop.c.  Before this change, the user would be prompted
  to confirm reloading, yet on confirmation reparsing from the source cache
  would be attempted.  So prompts were unnecessary and misleading -KW
* try to handle some failures of errors in source cache reparsing better.  Try
  to make sure that unsafe post_data documents don't get reloaded accidentally
  without confirmation if source cache reparsing fails.  Treat partial data
  (i.e.  loading from source cache probably 'z'apped by user) as successful, as
  HTLoadDocument does.  (Various error conditions are currently only passed on
  to the caller by HTParseFile, not by HTParseMem.  I left this apparent design
  decision unchanged.) -KW
* source cache reparsing did not rewrite a previous cache copy when the
  document was reloaded from the net.  This means that e.g. NOCACHE or RELOAD
  would force reloading from the network, but then on the next non-forced
  loading of the document (e.g.  by toggling to SOURCE view) the old (and
  possibly stale) copy would again be used.  Now the newer copy replaces the
  old copy if loading is successful (and not interrupted or aborted with 'z').
  (Note that RELOAD while in source view, at least in the normal one without
  -preparsed or -prettysrc, does not have this effect since the document does
  not get pushed through the cache writer at all in that case.) -KW
* if a source cache copy for a document exists, we need a way to ensure that
  reparsing from that copy does not try to rewrite that same copy "under our
  feet" while still reading from it.  Previously that was done by just never
  rewriting an existing source cache.  After changing that (see previous item),
  a different approach is needed to prevent such collisions.  This is now
  solved in two ways:  (a) Bse a temporary new file or memory chunk for
  writing, and only commit it to the doc's ParentAnchor object (replacing the
  previous copy) at the end of the incoming data stream.  This is done in
  CacheThru_free(), but not in CacheThru_abort(), meaning that an interrupted
  or otherwise aborted transmission will not replace an existing cached copy
  with a truncated now version - as long as such interruption gets detected and
  properly indicated, which is not the case for reading from the source cache
  with SOURCE_CACHE:MEMORY in effect, see above.  (b) Change the
  HTParentAnchor's protocol field to indicate a non-HTTP protocol, before
  starting to load from source cache in HTreparse_document().  Normally, as
  long as DEBUG_SOURCE_CACHE is not defined, this prevents the cache writer
  from (re)generating a cache copy.  Mechanism (b) isn't necessary (everything
  still works when DEBUG_SOURCE_CACHE is defined), it just avoids unnecessary
  work -KW
* source cache files and chunks were not always being removed when they should,
  probably since change of 1999-06-18.  A large number of cache copies could be
  kept around unnecessarily wasting disk or memory space.  Now make sure that
  the source cache copy gets removed when a document is removed from the cache
  of rendered docs.  They would never be reused anyway.  New function
  HTAnchor_clearSourceCache.  Note to other developers:  HTAnchor_delete()
  doesn't always remove the HTParentAnchor object, see its implementation -KW
* detect errors during appending to the cache copy.  Previously, memory
  allocation errors (for SOURCE_CACHE:MEMORY) would cause program exit, and
  file write errors (SOURCE_CACHE:FILE, e.g. if the temp disk space was full)
  would go undetected and leave the cache file corrupt or truncated.  Now allow
  the program to continue normally, just cancel the source cache generation for
  the current document (which may free up enough memory or disk space to remedy
  the immediate problem).  Produce an alert message.  In the case of a file
  error, produce the alert only once until a possibly re-reading of lynx.cfg,
  since Lynx should remain otherwise functional in the face of cache file
  writing errors -KW
* extended HTChunk: added a variant for which memory allocation errors
  are not fatal.  Used for changes described in previous item -KW
* don't write raw escape sequences to trace output for lynx-keymaps if
  not using trace log file -KW
* protect echo() in stop_curses, it could crash if stop_curses gets called
  without preceding start_curses -KW
* save size for regular files in anchor structure (we do the stat() anyway).
  It shows up on INFO page as Content-Length.  Also use it in partial display
  mode to prevent the last call to HTDisplayPartial from HTFileCopy, since a
  call to HText_pageDisplay will follow immediately.  (But note that nothing
  important depends on the correct size; should it be wrong, we lose at most
  one partial display screen update.  An equivalent suppression of the last
  partial update for network protocols is not recommended, since (a) the size
  is more unreliable and (b) the socket FIN may be delayed by the network.) -KW
* split HTMIME_put_character into three functions -KW
* made HTMIME.c handle folded header lines correctly (long-standing bug,
  reported in February by Devid Coles) -KW
  [ reference: http://www.flora.org/lynx-dev/html/month0299/msg00135.html ]
* changed how HTTP redirection messages are handled.  Don't read full message
  and parse buffer for some header fields in HTLoadHTTP.  Use common HTMIME.c
  header parsing instead.  Functions added/modified in HTMIME.c for storing
  away the Location URL (if requested and found) and for stopping after the
  headers.  New private MIME types "message/x-http-redirection" and "www/debug"
  to control this via HTStreamStack.  Basic approach taken from newer W3C
  libwww.  Benefits:
  - Removes actual bug that was reported: "Content-Location:" was falsely
    recognized as "Location:". (long-standing bug)
  [ reference: http://www.flora.org/lynx-dev/html/month1199/msg00370.html ]
  - Removes potential other bugs caused by inferior header parsing in HTTP.c
    (including Set-Cookie/Set-Cookie2).
  - Should more header fields be needed in the future (also) for redirection
    messages, we now get them automatically instead of having to add extra
    code.
  - Trace output now shows the same kind of info as for other responses.
    In particular, Set-Cookie/Set-Cookie2 header fields will normally be
    visible.
  - Statusline progress messages now actually show the HTTP response line
    as intended, and it is recorded for LYNXMESSAGES:. (Long-standing error,
    previously what was shown could be an empty string, or arbitrary parts
    of the response, including HTML markup from the message body.)
  - Adding redirection support for lynxcgi would now be simpler (although
    currently not done) -KW
* HTErrorStream(), similar to W3C libwww, used in some cases for the above -KW
* remove check in HTParse() dated 98/09/08 which prevented unescaping of 8-bit
  data (reported by KW)
* refresh screen for TRST changes in partial display mode as early as possible.
  This fixes a problem when the table finished below the screen bottom -LP
* remove global Newline_partial, use LYGetNewline() instead -LP
* adapt idea by VH as a Perl script cfg-html.pl which generates HTML form of
  the lynx.cfg documentation -TD