path: root/README

OpenSim (OS) and Second Life (SL) technology leave a lot to be desired. 
If they are ever going to take off, there needs to be a serious
redesign.  However, writing an entire virtual world system from scratch
is a huge job, especially for a tiny team.  So we will use the existing
OS and SL open source software as crutches, using them to run at least
one virtual world, while we slowly replace bits and pieces as we get to
them.  In this way, we can have our cake and eat it to.  Not having to
slave away for years just to get to the point where we have a single
avatar standing still in an empty world.

http://www.infinitegrid.org/drupal/content/OMG discusses our plans.  The
directory structure in here more or less follows the names of those
documents.

It's all very experimental at the moment, and currently does nothing
useful.


Design goals and methods.
=========================

As mentioned, this is all highly experimental.  Everything you find here
is subject to change without notice.  We do have some overall design
goals and methods though.

Simulating a world is what virtual world software does.  The real world
is full of lots and lots of stuff, and lots of that stuff has lots and
lots of fiddly little details to make it work.  So simulating a world
involves thousands, if not millions, of fiddly little details.

So yes you can have a working model of a car that you can actually get
inside and drive around a virtual world.  It doesn't have to have an
actually working internal combustion engine, GPS, radio, or air
conditioning.  The "engine" could just be a roughly engine shaped mesh
object with engine like textures wrapped around it to make it look like
a car engine.  Or the "engine" could just be a red cube on the front of
the car.  The "air conditioning" could just be a photo of a grill that's
part of the texture used for the "dash board".  Same for the "radio",
"GPS", or pretty much most of the other fiddly bits of a real car.  In
virtual worlds you can program a bare cube to act like a car that you
sit on and drive, the rest is mostly for looks.

There's no such thing as virtual wind in your hair, or smog in your
nostrils, but if that eventually gets invented, a generic virtual world
system should be able to hook into it easily enough.

Generic.
--------

The trick is to design generic stuff that can fake most if it well
enough to at least not annoy people.  Too much.

My experience is that if you do a good job of designing small useful
generic code, it becomes very easy to build things you never dreamed of
when you made the small generic bits.  It is also part of the Unix
philosophy, and you can do almost anything with the usual collection of
small Unix tools.

So the design of SledjHamr involves lots of little bits of generic code
that all work together seamlessly.  A major design goal is to be as
generic, yet useful, as possible, to support future stuff that hasn't
been invented yet.  As well as supporting stuff we need to do now.

If we do this right, then future work becomes really really easy. 
Virtual world development can then accelerate.

Small.
------

Small is the next major design goal.  For far to long people have argued
that computer storage, memory, and CPU power are all very cheap, so why
bother making things small.  Any long term computer user might have
noticed that despite our modern computers being many orders of magnitude
faster and bigger than they where a decade or two ago, everything runs
slower than it did then.  This proves the fallacy of "everything is cheap
and getting cheaper, lets be wasteful" theory.  People are just too
comfortable with their wasteful habits.  Not to mention that the
computer industry loves to get every one to throw out their computers
every couple of years and replace them with new shiny ones, just to make
big profits.  So every year normal software needs more powerful
computers, just to do what it did fine last year.

Even a small virtual world can use up huge amounts of storage, large
amounts of memory and CPU power, and large amounts of network bandwidth. 
This is the problem with simulating worlds, the bits might be small, but
there's an awful lot of them.  You can run a busy web site on a $5 per
month web host, but you need to spend at least $100 per month on a
hosted server that's powerful enough to run a small OpenSim based
virtual world.

The problem is scaling up to the level of detail needed.  So keeping
things small is an important design goal.  The generic design mentioned
above works for keeping things small, since you need less special
purpose code for all those special purpose fiddly details.

See the "Fast" section below about keeping network protocol and file
sizes small.

Fast.
-----

Ask any current virtual world user, at least of the types based on SL
technology, what the number one biggest problem is and they will tell
you it's lag.  So speed is important to everyone.  Which makes me wonder
why people use slow bloated things like interpreted scripting languages
and human readable network protocols / file formats for damn near
everything?

People use interpreted scripting languages coz it's easy and convenient
for them.  Not so convenient for the user though when it causes more
lag.  People invent human readable network protocols and file formats
coz it makes it easy for them to read when they need to debug the
system.  But 99.9999% of the time these human readable network and file
formats are NEVER READ BY HUMANS.  Computers don't find them easy to
read at all.  It takes more effort for the computer to figure it out,
more space on your hard drive, more space in your memory, and more
bandwidth to send.  All they do is slow everything down massively most
of the time.

So basically a small handful of developers are taking the easy way out,
but that inconveniences the vast majority of people.  Mind you, you cant
always blame the developers, money is involved, which means people are
forced by management to take what ever shortcuts they can now, even if
it means pain later.  That's economics for you.

We wont follow that bullshit here.

Once again it's a matter of scale for virtual worlds.  Each part by
itself might be barely fast enough that people don't notice, but it all
adds up when you deal with the huge multitude of fiddly little details
in a virtual world.  Which results in the number one problem being ...
lag.  Often everything slows down so much it becomes unusable.

We can have our cake and eat it to.  So long as the crucial heavily used
parts of the system that need speed are written efficiently in a
decently fast language, then we can still use easy interpreted
scripting languages for other parts that wont suffer from scaling
issues.  So long as we stick with efficient binary based network
protocols and file formats, the tiny percentage of developers that need
to actually read it can use specialised tools for that job that turn
them into human readable forms, but not slow down every body else.

It's my experience, and that of other people that work with code that
needs to be low level, fast, and small, that the C language, with help
from assembler sometimes, is the best choice for fast working code. 
Assembler and C is used for OS kernels, embedded software, and other
things that need to be efficient.  So C is the major language used for
SledjHamr.  Bits of assembler might be used if needed.

For the interpreted scripting language of choice used in SledjHamr, I
chose Lua.  It's very generic in nature with it's wonderful tables and
meta tables.  It's tiny, designed to be embedded inside other languages. 
With the LuaJIT just in time compiler, it's the fastest scripting
language around.  These three things fit our major design goals.  On top
of those things, it's popular with other 3D online environments,
especially WoW.  It's also very easy to use and can be quite fast for
development, which is why you use a scripting language in the first
place.  The ability to be embedded in other code means that Lua tends to
actually be embedded in many other projects.  Which can then use Lua to
build all manner of oddball mash ups.  Exactly what virtual worlds need.

While on the subject of languages, see the stuff below about EFL.  Also,
when Apple based their new version of Mac Os X on BSD Unix, the OS wars
where over.  Everything now is some sort of Unix, except Windows.  Hell,
iPhones OS is based on Mac OS X, and Android is based on Linux, so even
the great majority of phones these days are Unix.  Every one is a Unix
user, whether they know it or not.

Windows makes a nod to being Posix compliant, though it's barely a nod. 
Cygwin however can be installed on Windows to make it more like the Unix
used in the rest of the world.  So in general Posix and Unix style stuff
should be used.  Try to keep the Windows specific stuff well isolated,
small, and preferably based on Cygwin.  This should lead to A) maximum
portability.  B) Less stress for the SledjHamr head developer, who got
burnt out working with Windows crap long ago and really really REALLY
wants to avoid it as much as possible now.  The OS wars are over, Unix
won, fucking get over it Microsoft.

Modular.
--------

This is actually part of being generic and small.  If everything is a
small module, then you only need to use those modules that do the things
you are doing right now.  The rest you don't need, so you don't load
them up.  No need to cater for all of those fiddly little details in one
massive world simulating blob, break it down into small modular and
generic bits that talk to each other using common protocols and
standards.

This can also be useful to support the great variety of other file
formats and protocols already being used.  Need to support some obscure
file format no one has heard of?  Write a small module for it.  All the
major file formats can be supported out of the box.  Want to join up
with an obscure virtual world platform?  Write a module to support it. 
All the common virtual world tasks will be already supported by
SledjHamr, you just need to translate between the file format or network
protocol in your module.  OpenSim, SL, and later Cobalt, will be the
initially supported virtual world network protocols.

For the purposes of using the old SL type technology as a crutch during
development, being modular makes a lot of sense.  When any given module
is good enough to start using, we can hook it up to Imprudence or
OpenSim to start using it for real.  When that module gets mature, we
can rip the relevant code out of those older systems.  Eventually those
older code bases morph into SledjHamr based ones.

Standards based.
----------------

OK, we will cheat a bit here.  No such thing as a decently small and
fast virtual world protocol standard, so we are inventing our own.

On the other hand, we intend to use open standards based network
protocols (even if they are bloated human readable ones) to take care of
the various bits not specifically related to virtual worlds.  Like
Jabber/XMMP for text communications, HTTP for asset transfers, etc. 
These protocols are well used, well understood, and in general not TOO
bloated for their intended use cases.  These use cases are close enough
to our own that we can leverage this huge installed base of code and
expertise to deal with the problems that virtual worlds would share. 
Much less work for us all.  People can use the software they are used to
to interface with those bits.

Standard texture file formats tend to be very well compressed, and the
libraries for them more or less mature.  Lots of research and
standardisation effort has gone into the major popular ones.  So we
don't really need to deal with that.  Problem solved already.  Well,
except for JPEG 2000, which outside of SL and medical use, no one has
even heard of.  It's library support is not so good, and made even worse
by the fact that the best decoding algorithm for it is patented and out
of reach to open source developers.  Still, lots and lots of SL and
OpenSim content is in JPEG 2000 format, and it is good as a network
texture format, so we just have to deal with it.

Mesh file formats are a bit less standardised, and a bit less well
compressed, but the libg3d mesh loading library deals with that well
enough, and it's good to be able to use all that content out there.  On
the other hand, OpenSim OAR and IAR files are some what common, but best
to use just for importing OpenSim based worlds into SledjHamr.  They are
not pretty on the inside.  On the gripping hand, SL viewers never really
standardised on prim file formats.  In general you need to use the
viewer you used to export to import them.  However, this is the sort of
thing the modularity is made for.  Write small modules to deal with
these SL inspired file formats.  Make them libg3d modules.

OpenGL is the worlds open 3D standard, and even works on Windows.  It's
well known that Microsoft (at least in the past) made their OpenGL
implementation slower than their own 3D system, specifically so they
could claim theirs was better.  OpenGL works fine on Windows, lets use
it.  Any slowness issues we just blame on Microsoft.

Secure.
-------

One of our developers is a cypherpunk / cryptogeek / whatever term she's
comfortable being labeled as.  Our team is very privacy focused.  So
security and privacy are important goals as well.  Small modular code is
better for security, is there is less code to look at to do security
audits, and less places for things to go wrong, or escape attention. 
Commonly used security and privacy protocols also tend to be built in
modular ways, so we can make use of these things ourselves.

Pretty and skinable.
--------------------

The Enlightenment Foundation Libraries (EFL) where written to support
the Enlightenment X window manager, DR17 (E17).  Enlightenment is known for
it's eye candy.  The early DR16 version was popular, but known to be a
heavy resource user.  That was many years ago, and the most popular X
window managers these days far surpassed Enlightenment DR16 for bloat,
without coming close for eye candy.

E17 was rewritten from scratch, and I helped with that rewrite in
various ways.  E17 is a lot less "bloated", but I think that's mainly
coz everything else passed it in bloatedness, even though much work was
done to ensure E17 would work great on small embedded systems.  I have
used EFL on professional embedded systems, and it worked well, even on
the tiny x486 embedded board I used.  In general EFL fits well with my
personal coding standards, and fits well with the SledjHamr goals.  This
is why I continue to work with it, and used it for SledjHamr.

Basically, you can think of EFL as a replacement for the GTK that SL
viewers use.  They both offer more or less a similar feature set, and
both use more or less the same standard libraries underneath.  GTK uses
the GLib main loop, which EFL supports via wrappers.  I think that EFL
fits the SledjHamr design goals much better though.

As a bonus, lots and lots of eye candy, complete with a system for
writing your own eye candy, and over riding the system supplied eye
candy.  So very pretty GUI skins are supported.  As a special bonus,
works great on small systems.  EFL is being used for Samsungs Tizen, in
fact half the EFL developers now work for Samsung.

Flexible.
---------

The generic and modular qualities mean that people can build their own
virtual worlds in any way they like.  Use a file based asset store, or a
database based one, or generate it on the fly using genetic algorithms,
what ever floats your boat.  Serve assets with Apache, Nginx, lighttp,
or any other web server.  Use ejabberd, or the Lua based Prosody for
text based communications.  Maybe you want to use a CMS like Drupal to
deal with assets, users, groups, and permissions?  Let your users
manipulate their inventory, or the world, using WebDav.  Use LDAP for
users and groups.  Roll your own if you don't like any of the other
choices for any specific module.  SledjHamr wont stand in your way, and
will help you use the virtual world specific bits any way you want.

Users get this flexibility to.  Use GIMP or Photoshop to edit textures
directly.  Or Blender to create new meshes in world.  Keep your
inventory on your own hard drive, so you can use the tools you are
comfortable with to manage it all.  Store your inventory in "the cloud". 
Easily send things from your inventory to others without even logging in
world, using what ever method you like.  Share some of your inventory on
social media.

Create what ever sort of mash up you can bend your tools to do.  Hook up
an IRC bot to an in world group chat room.  Run on online business that
prints in world objects on 3D printers, using standard E business
software and a virtual world as your shop front end.

The virtual world is your oyster, go nuts.

Er, but experimental for now.
-----------------------------

Still early days, and even though a few helped with the first design
docs, so far all the coding work has been done by one person.  So much
to do for a virtual world system.  This is why the git repo has an empty
master branch, it's all in the experimental branch.  So far only bits
and pieces, which have only recently been tied together into one system.

As experimental software, it's all up in the air as we try out different
things, explode our own careless thought bubbles, rip things out and
start from scratch, and in general mess with things to our hearts
content.  Nothing is set in stone yet.  We write some small bit of code
that seems at least slightly sane, or cobble something together quickly
that might be totally insane.  Then tinker with it to see what works, or
to make it closer to our goals.  Eventually we will make bits that we
think are stable enough for actual use and move them to the master
branch.  Don't go holding your breath though.

Lua has been great for quickly building prototype stuff.  So far most of
the network protocols and file formats have been based on Lua.  Which
goes against our goals, so expect that to be temporary.  Specifically,
once we get around to actually implementing the nails protocol stuff,
some of the Lua network and file formats will get replaced by nails.

The Lua GUI code is loosely based on my ancient matrix-RAD stuff that was
written in Java.  The fact that I had most of it working within a week
is a tribute to how easy Lua is.  The original Java took me years to
write.  On the other hand, I'm not that happy with the syntax of the
results, so that is likely to change.

Other developers could use their own favourite languages for
experimental work, but I highly encourage them to eventually produce C
or Lua code depending on the requirements of that module.

The specific coding style used is all over the place at the moment. 
Lots of code I wrote for other projects has been dragged in and morphed
into our requirements, but it all came with it's own style from those
projects.  I've produced an uncrustify file for cleaning up C code, but
not applied it to most code yet.  Even that uncrustify file is subject
to tweaking.

The LuaSL code is the most stable, but that's not surprising, it's the
first thing I wrote.  On the other hand, all of the actual world
changing LL function calls have not been written yet, there's no actual
world code to put them in.  Some one volunteered to interface LuaSL to
OpenSim, which would pass the world changing functions to that to deal
with, but they vanished before they started.  LuaSL at least uses a
network socket to feed the world changing functions to a world server,
so that part can be modular and support OpenSim, the SledjHamr love
server, or anything else.

I've not settled on which 3D rendering system to use yet.  The options I
have short listed and experimented with are Irrlicht and Evas_3D.

I had looked at various open source 3D render systems, including things
like Orge, with a specific set of goals in mind.  A major goal was to
support the plethora of low powered laptops every one uses for virtual
worlds that have bare minimum 3D support, or even none at all.  Irrlicht
has a great software renderer for people with no 3D hardware.  That got
my vote, so I experimented with it.  There was no EFL integration, so I
had to write some, which wasn't that hard.  A year later, a new version
of EFL managed to bit rot my EFL/Irrlicht integration, so I fixed that. 
But now it flickers like crazy.  After much discussion with other EFL
developers, and some preliminary work by the Irrlicht developers, an
experimental version of Irrlicht is underway that might integrate better
with EFL.  I've not tried it yet, but it was on my TODO.

In the mean time, and so recently that it only gets released later
today, Evas_3D was added to EFL.  Evas is one of the sub libraries of
EFL that deals with canvasses. Evas_3D is a 3D mesh, camera, and light
system designed to be part of Evas.  So it has the major advantage of
already being well integrated, it's part of EFL.  I've added test code
for it now, and no flickering.  I even have test code using both
Irrlicht and Evas_3D in the same window space.  It's major disadvantage
is that it's only a very recent project that has (until later today) not
been released, Irrlicht is much more mature.  So Evas_3D only has basic
stuff, though it's mostly complete basic stuff.  On the other hand, a
lot of what's missing in Evas_3D I was thinking about rewriting for
Irrlicht anyway.  So far one major missing bit is Bullet Physics
integration, or any other physics engine.  Irrlicht has Bullet, and
another part of EFL also has Bullet.  I'm hoping the authors of that
other EFL part get together with the Evas_3D authors and figure
something out.

So right now I'm leaning more in the Evas_3D direction, and that's
currently the default.  In neither case had I written anything to
actually create a proper virtual world, merely modifications to the
examples provided to see how they fit in to the rest, and give me
something to play with.  I think I'll see how far I can get with Evas_3D
before making a decision.

As I was writing the above stuff about Evas_3d being released later
today, it got released.  lol



The bits (or some of them).
===========================

extantz
    GuiLua front end for other programs it starts
    3D rendering of multiple virtual worlds in tabs
	    including the same virtual world logged in more than once with different accounts
	    or even just different camera views
	    could use the same thing for generating two views for stereo viewing.
    Nails interface to virtual world backend modules.  Each module converts nails commands to / from it's own network protocol.
	A SledjHamr grid, which definitely should be running independently, 
	    so others can log on and stay on when extantz closes down
	    which may be a local or remote grid
	    might be part of some other grid (a sim)
	A local only SledjHamr grid, with no other logins.
	    loaded from an on disk SledjHamr grid, or even across the 'net
	    which can be persisted as a SledjHamr grid on disk
	An OS/SL module that connects up to OS/SL grids and translates between OS/SL networking protocols and nails / GuiLua.
	    sending OS/SL network crap to the grid in response to nails commands
	    getting OS/SL network crap from the grid
		translating and sending nails commands to extantz so it can update it's 3D models
		  including nails encapsulated GuiLUa commands to deal the with OS/SL GUI stuff
	An Open Croquet / Cobalt module.
    importer / exporter
	Which basically grabs an area / selected objects from the current tabs world,
	then sends them as nails commands to -
	    a file
	    across the 'net
	    the inventory browser
	    the IAR & OAR file writer
	or grabs one of those things listed before, and sends it as nails commands to the current tabs world

GuiLua
    library of all the matrix-RAD type stuffs, so others can use it
    should include the code so that proggies using it can be client, server, or both
    including the GuiLua script runner
	Should only be running dozens at most, and at human speeds (maybe).
	In the end Lua is lean on resources by design, so we can afford to have lots of proggies using it all over.
    matrix-RAD style -
	server creates skang GUI, sends it to client
	client deals with GUI stuff, sends values back to server
	server stores values and does things
	client can do things to

Edje Lua
    Lua embedded in edje files, and sandboxed to them.
    add table marshaling into an edje message
    add host proggy supplied Lua functions
	So we can add nails.foo(), GuiLua.foo(), and maybe even LSL.foo().
	    All users of GuiLua and nails probably want to be sandboxed, the scripts should be loaded up by extantz, OpenSim, or SledjHamr, not run from the operating system.
	LuaSL adds these to their generated scripts -
	    local _bit = require("bit")
	    local _LSL = require("LSL")
	But require() is part of the package library that Edje Lua disables.
	    Tweak Edje Lua so that we can use it for external GuiLua and LuaSL scripts.
		Edje Lua can add the bit library if it detects LuaJIT or 5.2.
		LuaSL stuff needs to be sandboxed as well, maybe we can use this new host function thingy to add LSL to?
		Really should investigate if this shit being duplicated in thousands of LuaSL scripts soaks up lots of memory?  And require("LSL") versus this new host function thingy.
		On the other hand, a lot of this LSL library would end up being wrappers around nails in the future, which is a shared library.
    LuaJIT?  Optional?
	Need to sort out it's memory allocator, coz apparently the one Edje uses is not thread safe, but the LuaJIT one does not limit memory per script like Edje does.
	    Maybe Edje can use an Eina allocator?
	    LuaSL uses luaL_newstate() for now, which uses the default Lua allocator (which might be the LuaJIT one if that's in use).
	    Edje uses lua_newstate(_elua_alloc, &ela) instead.
    merge the actual script threading code from LuaSL into Edje Lua?
	once this code is in Edje Lua, both LuaSL and GuiLua can use that

woMan
    account manager
	GuiLua for dealing with the account info
	either sending the GuiLua to extantz, or dealing with it itself
    settings manager
	reading legacy viewer XML GUI files for preferences
	reading legacy viewer XML settings files
	translating that to GuiLua
	either sending the GuiLua to extantz, or dealing with it itself
	writing changed settings to legacy viewer XML settings files

in world object editor
    Good to be separate, coz often you spend a lot of time NOT editing.
    GuiLua based, so it can even be replaced / hacked up / tweaked.
    Probably not much point dealing with GUI itself, as you want a 3D rendering of what you are editing, thus need extantz as the front end.
    sending nails to extantz to render the results
	extantz sends nails to the virtual world backend modules, which deal with translating and sending them on
	also gotta deal with OS/SL backends that prefer to send the entire thing round robin and only update the screen once it's hit the server

inventory browser
    Good to be separate, coz often you spend a lot of time NOT dealing with inventory.
	On the other hand, people might have their inventory window open all the time, I do.  lol
	Naturaly you could be dealing with inventory while NOT being logged in world.
    handles local inventory as the usual nails files

LuaSL
    socket for other programs to send scripts and other info through
    LSL -> LuaSL compiler
    LuaSL script runner - likely to be running thousands of scripts
	LuaJIT to speed it up, EFL luaproc to run them as worker threads
	Has since been rewritten to be an Ecore_Thread  / cooperative multitasking based system, depending on compile options.
	  Should make that choosable at run time after more testing.
    LSL constants and functions implemented as a Lua library
	The functions want to use nails to talk to the server backend.
	Currently just sending Lua function() call strings instead.
	The "server backend" could be OpenSim (with a shim), an actual SledjHamr server (direct nails), or extantz (direct nails).
	The server backend is whatever is on the other end of the socket.
	    it should be able to deal with multiple socket users, but may have to fix it if it does not
	LuaSL scripts might include LSL style GUI elements (very few, very limited)
	    convert them to and from nails encapsulated GuiLua
		if what's on the other end of the socket is extantz, no worries
		otherwise the other end is OpenSim, the OS end can translate GuiLua into OS/SL viewer network stuff
		    let the OS end deal with the issues.  It's their fault.  :-P
	LuaSL (but not LSL) scripts could include GuiLua and other nails stuff directly.
	    Note, OpenSim being the server backend wont know what to do with GuiLua, and will only know about nails later.
		This is just a temporary thingy anyway.
    Might be able to convert it to use more generalised message parsing, PLUS teach Edje some of LuaSL's message parsing tricks.  Find a happy medium.

nails command pump
    library, including C and Lua bindings, so all can use it as client, or server, or both
    dedicate one nails command to encapsulate GuiLua skang commands
	"G", so GuiLua skang can look like "GuiLua.widget("name", "Button", 0, _0, 100, 4, 0, "label");"
    could even have a nails command for Lua / LuaSL / LSL commands or scripts.  This might help with the OpenSim nails shim, we can start with that, and expand it later.  B-)
	"L" and "l", though currently LuaSL is using SID.foo(), where SID is the UUID of the script.  Easy enough to whack an L in front of that.
	The LuaSL scripts are using _LSL as the table name, but that's just using _ as an "internal variable" marker.  Also easy to change.
    SledHamr grids / sims are stored on disk or sent across the 'net as nails commands.

IM client
    Separate coz it could also be used as a normal IM client.
    either sending the GuiLua to extantz, or dealing with it itself
    calls libpurple for the hard work
      Or maybe not, I've gone off libpurple, even after naming the IM stuff purkle.
    needs stuff for extantz / woMan to feed it IM / group / local chat stuffs from grids
	gotta be modular, perhaps same module that handles other grid stuff for extantz?
	or a libpurple module for OS/SL, and have authentication credentials for those grids passed from elsewhere (woMan?)

web browser
    We may have to deal with external web proggies, which may in the end be the best way.
    Except for MOAP.  lol
    Maybe later we can have a HTML to GuiLua translater module?  Muahahaha

IAR & OAR file reader / writer
    loads up the file, sending it as nails commands
    converts and stores incoming nails commands as an IAR or OAR file
    Could cover Impy exports as well as others?