There are a few problems that are reported by some users, and which are not that easy to fix for specific reasons explained here. In the tradition of other software makers, I'll call these problems to be problematic features, not bugs.
In the traditional pfe and in the case of an ABORT (or any other THROW that is not CATCHed), the simple approach was use that the current input-buffer was printed and a single "^" did mark the current position of the ">IN" pointer, which is the index within the input-buffer where the next PARSE will start off.
However, the gforth showed that it is better for the user to see the last WORD being PARSEd last from the input - that is, the whole last charlenght shall be underlined. That is almost simple since there is an internal SPAN variable in forth that gives us the length of the last PARSE value, and this SPAN variable is not transient (temporary in a way).
Now, what's the problem - the outer interpreter will PARSE the next word from the input buffer using a BL space. The PARSE will read from the input buffer until it hits either that space OR the end of the input buffer. If there is a space then the ">IN" index will be left pointing to the point one beyond that space, that is two chars beyond the last valid character of the WORD just parsed out. However, if the PARSE hits the end of the input-area, the ">IN" pointer will be just after the end of the input-area, that is one char beyond the last valid character of the word just parsed out.
Now, how to detect the difference - the difference whether an additional delimiter has been parsed away that was not seen by the word calling PARSE (or any other special purpose parsing function) and which therefore does not add to the SPAN returned. The answer is that it is very hard - we could add the heuristic that at the end of the input buffer we just add one, but there are very different input SOURCEs each having a different idea of the actual end (think of a block file that is read in one go), and it we would have to check the buffer whether there might be a single delimiter just before the actual end.
So actually, the real answer would be to place additional variables to the parsing functions that can be checked at the next THROW. However, it simply means to slow down the parsing functions a bit, and all of them must follow this standard to publish what they've been returning from the input buffer.
Here I took the decision to not add either of additional complexity in the internal parsing or execption facilities. Instead, (at an ABORT) the underline of the last PARSE simply shows SPAN+1 characters - inside the input buffer, that will include the delimiter just after the last WORD, and if we had been at the end of the input buffer, it will underline the word plus the delimiter just before the last WORD. That is good enough to give the user an idea what characters in the input buffer had been affected in the last PARSE - a human mind can easily subtract addtional delimiters before and after such a word, which is usually a space or a doublequote.
In the end, the PFE does now show the complete SPAN of the last PARSEd word without adding much of additional algorithmic complexity to the forth engine. That keeps the PFE code readable and maintainable at the cost of users who sometimes wonder why the space after word is underlined too. Well, now you know...
There is a difference between an include-file that is given as the bootscript to the pfe binary via the shell commandline, and the loading of the same file via "INCLUDE >file<" directly from the forth commandline. This is most obvious with the ORDER that is forgotten when being set in a bootscript as soon as it hits the the mainloop - when the bootscript has finished, the inner endless loop is setup and will run first through the QUIT-word and ABORT-word initialization sequences which includes a RESET-ORDER.
To give you an idea, here is what is going on: first the the forth dictionary is allocated, then it gets initialized from the wordsets being compiled into the main forth binary, then the bootimage is loaded, then the bootscript is loaded, then we look for APPLICATION and jump into it. The default APPLICATION is the forth-internal MAINLOOP. The mainloop does build a CATCH-domain around the traditional forth-routine QUIT. Calling QUIT will first go through some initializations that will ensure that the forth machine is in a sane state even when it is run again explictly or implictly from an ABORT that did end in the CATCH-domain of the MAINLOOP.
That's why you can not leave values from the bootscript on the parameter stack - they will be lost. There can not be any files left open from the bootscript either, they will be closed automatically. While that does not hurt many people, the case about the order-reset can be problematic in some cases as it might change the parsing behaviour of the outer interpreter as soon as the mainloop's QUIT is reached. To get around this, you can use the PFE-specific word DEFAULT-ORDER to explictly set the order that RESET-ORDER will see.
Shall we implicitly do a DEFAULT-ORDER after loading the bootscript? That might help us but in the real world there are far too many forth scripts that change the ORDER via a lot of "ALSO" additions starting off with an "ONLY FORTH" but they never call any "PREVIOUS" at the end as they expect that the next module script will do the same and start off at "ONLY FORTH". Furthermore, there is a basic differene between a bootscript and a normal include-file since the APPLICATION variable is only checked after the bootscript but not after each include-file. In a way the include-script on the shell commandline will not be loaded through INCLUDE but through BOOTFROM which builds on top of INCLUDE.
Just make sure that an application bootscript is ready to take this difference - one of the differences is also that any error in a bootscript will let the PFE instance simply die since there is no CATCH-domain inside an endless loop, and it is up to the APPLICATION to create a CATCH-domain and a PARSE-loop if needed. It could be that a forth program is only used as a filter for some input from the shell commandline so that it shall never reach the forth commandline. That is up to the bootscript to flag (well, you can also use the comandline option "-y"/"--bye" to let APPLICATION point to BYE instead of MAINLOOP).
After all - the forthscript on the shell commandline is not exactly the same as loading the same file from within the forth commandline. In general, there should be no difference, in other cases it might lead to some unexpected differences. Now you know why...
P.S. in the Tek/MPT embedded variant, there is no use for a PFE that quits immediatly, in fact it doesn't know BYE, it can just do a COLD reboot. Here the bootscript is sent to the PFE process via putting the resp. forth commands into the terminal input queue, so it is seen within the endless-loop, and all scripts will make use of the PFE-specific DEFAULT-ORDER word to ensure a good state at an ABORT.
COLD does not reset ORDER ... A problem that existed since a change in 0.32.x and which should go away. It has to do with COLD going not deep enough due to the bootscript mechanics vs include-files.
Call-threading has no debugger ... new to 0.32.x are some more threading modes including SBR-threading. While the latter is largely unstable and not available on all platforms, the CTC threading however is completly portable and should be supported by a debugger too.