Mercurial > cgi-bin > hgweb.cgi > tincan
view tincan.py @ 53:3a6180c1feea draft
Remove some needless repetition.
author | David Barts <n5jrn@me.com> |
---|---|
date | Fri, 31 May 2019 08:23:37 -0700 |
parents | 997d0c8c174f |
children | cb5a6e200c95 |
line wrap: on
line source
#!/usr/bin/env python3 # -*- coding: utf-8 -*- # As with Bottle, it's all in one big, ugly file. For now. # I m p o r t s import os, sys import ast import binascii from base64 import b16encode, b16decode import email.utils import functools import importlib from inspect import isclass import io import logging import mimetypes import py_compile from stat import S_ISDIR, S_ISREG from string import whitespace from threading import Lock import time import urllib import bottle # E x c e p t i o n s class TinCanException(Exception): """ The parent class of all exceptions we raise. """ pass class TemplateHeaderError(TinCanException): """ Raised upon encountering a syntax error in the template headers. """ def __init__(self, message, line): super().__init__(message, line) self.message = message self.line = line def __str__(self): return "line {0}: {1}".format(self.line, self.message) class LoadError(TinCanException): """ Raised when we run into problems #load'ing something, usually because it doesn't exist. """ def __init__(self, message, source): super().__init__(message, source) self.message = message self.source = source def __str__(self): return "{0}: #load error: {1}".format(self.source, self.message) class ForwardException(TinCanException): """ Raised to effect the flow control needed to do a forward (server-side redirect). It is ugly to do this, but other Python frameworks do and there seems to be no good alternative. """ def __init__(self, target): self.target = target class TinCanError(TinCanException): """ General-purpose exception thrown by TinCan when things go wrong, often when attempting to launch webapps. """ pass # T e m p l a t e s # # Template (.pspx) files. These are standard templates for a supported # template engine, but with an optional set of header lines that begin # with '#'. class TemplateFile(object): """ Parse a template file into a header part and the body part. The header is always a leading set of lines, each starting with '#', that is of the same format regardless of the template body. The template body varies depending on the selected templating engine. The body part has each header line replaced by a blank line. This preserves the overall line numbering when processing the body. The added newlines are normally stripped out before the rendered page is sent back to the client. """ _END = "#end" _LEND = len(_END) _WS = set(whitespace) def __init__(self, raw, encoding='utf-8'): if isinstance(raw, io.TextIOBase): self._do_init(raw) elif isinstance(raw, str): with open(raw, "r", encoding=encoding) as fp: self._do_init(fp) else: raise TypeError("Expecting a string or Text I/O object.") def _do_init(self, fp): self._hbuf = [] self._bbuf = [] self._state = self._header while True: line = fp.readline() if line == '': break self._state(line) self.header = ''.join(self._hbuf) self.body = ''.join(self._bbuf) def _header(self, line): if not line.startswith('#'): self._state = self._body self._state(line) return if line.startswith(self._END) and (len(line) == self._LEND or line[self._LEND] in self._WS): self._state = self._body self._hbuf.append(line) self._bbuf.append("\n") def _body(self, line): self._bbuf.append(line) class TemplateHeader(object): """ Parses and represents a set of header lines. """ _NAMES = [ "errors", "forward", "methods", "python", "template" ] _FNAMES = [ "hidden" ] _ANAMES = [ "load" ] def __init__(self, string): # Initialize our state for i in self._NAMES: setattr(self, i, None) for i in self._FNAMES: setattr(self, i, False) for i in self._ANAMES: setattr(self, i, []) # Parse the string count = 0 nameset = set(self._NAMES + self._FNAMES + self._ANAMES) seen = set() lines = string.split("\n") if lines and lines[-1] == "": del lines[-1] for line in lines: # Get line count += 1 if not line.startswith("#"): raise TemplateHeaderError("Does not start with '#'.", count) try: rna, rpa = line.split(maxsplit=1) except ValueError: rna = line.rstrip() rpa = None # Get name, ignoring remarks. name = rna[1:] if name == "rem": continue if name == "end": break if name not in nameset: raise TemplateHeaderError("Invalid directive: {0!r}".format(rna), count) if name in seen: raise TemplateHeaderError("Duplicate {0!r} directive.".format(rna), count) if name not in self._ANAMES: seen.add(name) # Flags if name in self._FNAMES: setattr(self, name, True) continue # Get parameter if rpa is None: raise TemplateHeaderError("Missing parameter.", count) param = rpa.strip() for i in [ "'", '"']: if param.startswith(i) and param.endswith(i): param = ast.literal_eval(param) break # Update this object if name in self._ANAMES: getattr(self, name).append(param) else: setattr(self, name, param) # C h a m e l e o n # # Support for Chameleon templates (the kind TinCan uses). class ChameleonTemplate(bottle.BaseTemplate): def prepare(self, **options): from chameleon import PageTemplate, PageTemplateFile if self.source: self.tpl = PageTemplate(self.source, **options) else: self.tpl = PageTemplateFile(self.filename, encoding=self.encoding, search_path=self.lookup, **options) # XXX - work around broken Chameleon decoding self.tpl.default_encoding = self.encoding def render(self, *args, **kwargs): for dictarg in args: kwargs.update(dictarg) _defaults = self.defaults.copy() _defaults.update(kwargs) return self.tpl.render(**_defaults) chameleon_template = functools.partial(bottle.template, template_adapter=ChameleonTemplate) chameleon_view = functools.partial(bottle.view, template_adapter=ChameleonTemplate) # U t i l i t i e s def _normpath(base, unsplit): """ Split, normalize and ensure a possibly relative path is absolute. First argument is a list of directory names, defining a base. Second argument is a string, which may either be relative to that base, or absolute. Only '/' is supported as a separator. """ scratch = unsplit.strip('/').split('/') if not unsplit.startswith('/'): scratch = base + scratch ret = [] for i in scratch: if i == '.': continue if i == '..': ret.pop() # may raise IndexError continue ret.append(i) return ret def _mangle(string): """ Turn a possibly troublesome identifier into a mangled one. """ first = True ret = [] for ch in string: if ch == '_' or not (ch if first else "x" + ch).isidentifier(): ret.append('_') ret.append(b16encode(ch.encode("utf-8")).decode("us-ascii")) else: ret.append(ch) first = False return ''.join(ret) # The TinCan class. Simply a Bottle webapp that contains a forward method, so # the code-behind can call request.app.forward(). class TinCan(bottle.Bottle): def forward(self, target): """ Forward this request to the specified target route. """ source = bottle.request.environ['PATH_INFO'] base = source.strip('/').split('/')[:-1] if bottle.request.environ.get(_FTYPE, False): raise TinCanError("{0}: forward from error page".format(source)) try: exc = ForwardException('/' + '/'.join(_normpath(base, target))) except IndexError as e: raise TinCanError("{0}: invalid forward to {1!r}".format(source, target)) from e raise exc # C o d e B e h i n d # # Represents the code-behind of one of our pages. This gets subclassed, of # course. class BasePage(object): """ The parent class of both error and normal pages' code-behind. """ def handle(self): """ This is the entry point for the code-behind logic. It is intended to be overridden. """ pass def export(self): """ Export template variables. The default behavior is to export all non-hidden non-callables that don't start with an underscore. This method can be overridden if a different behavior is desired. It should always return a dict or dict-like object. """ ret = { 'page': self } for name in dir(self): if name in self._HIDDEN or name.startswith('_'): continue value = getattr(self, name) if callable(value): continue ret[name] = value return ret class Page(BasePage): """ The code-behind for a normal page. """ # Non-private things we refuse to export anyhow. _HIDDEN = set([ "request", "response" ]) def __init__(self, req, resp): """ Constructor. This is a lightweight operation. """ self.request = req # app context is request.app in Bottle self.response = resp class ErrorPage(BasePage): """ The code-behind for an error page. """ _HIDDEN = set() def __init__(self, req, err): """ Constructor. This is a lightweight operation. """ self.request = req self.error = err # I n c l u s i o n # # Most processing is in the TinCanRoute class; this just interprets and # represents arguments to the #load header directive. class _LoadedFile(object): def __init__(self, raw): if raw.startswith('<') and raw.endswith('>'): raw = raw[1:-1] self.in_lib = True else: self.in_lib = False equals = raw.find('=') if equals < 0: self.vname = os.path.splitext(os.path.basename(raw))[0] self.fname = raw else: self.vname = raw[:equals] self.fname = raw[equals+1:] if self.vname == "": raise ValueError("empty variable name") if self.fname == "": raise ValueError("empty file name") if not self.fname.endswith(_IEXTEN): raise ValueError("file does not end in {0}".format(_IEXTEN)) # Using a cache is likely to help efficiency a lot, since many pages # will typically #load the same standard stuff. Except if we're # multithreading, then we want each page's templates to be private. _tcache = {} def _get_template_cache(name, direct, coding): aname = os.path.abspath(os.path.join(direct, name)) if aname not in _tcache: tmpl = ChameleonTemplate(name=name, lookup=[direct], encoding=coding) assert aname == tmpl.filename _tcache[aname] = tmpl return _tcache[aname] def _get_template_nocache(name, direct, coding): return ChameleonTemplate(name=name, lookup=[direct], encoding=coding) # R o u t e s # # Represents a route in TinCan. Our launcher creates these on-the-fly based # on the files it finds. _ERRMIN = 400 _ERRMAX = 599 _IEXTEN = ".pt" _PEXTEN = ".py" _TEXTEN = ".pspx" _FLOOP = "tincan.forwards" _FORIG = "tincan.origin" _FTYPE = "tincan.iserror" class _TinCanBaseRoute(object): """ The base class for all NON ERROR routes. Error routes are just a little bit different. """ def __init__(self, launcher, name, subdir): global _get_template_cache, _get_template_nocache if launcher.multithread: self.lock = Lock() self.get_template = _get_template_nocache else: self.lock = _DummyLock() self.get_template = _get_template_cache self.logger = launcher.logger def urljoin(self, *args): """ Normalize a parsed-out URL fragment. """ args = list(args) if args[0] == '/': args[0] = '' return '/'.join(args) def launch(self): raise NotImplementedError("This must be overridden.") def __call__(self): raise NotImplementedError("This must be overridden.") class _TinCanStaticRoute(_TinCanBaseRoute): """ A route to a static file. These are useful for test servers. For production servers, one is better off using a real web server and a WSGI plugin, and having that handle static files. Much of this logic is cribbed from the Bottle source code (we don't call it directly because it is undocumented and thus subject to change). """ def __init__(self, launcher, name, subdir): super().__init__(launcher, name, subdir) self._app = launcher.app self._fspath = os.path.join(launcher.fsroot, *subdir, name) self._urlpath = self.urljoin(launcher.urlroot, *subdir, name) self._type = mimetypes.guess_type(name)[0] if self._type is None: self._type = "application/octet-stream" if self._type.startswith("text/"): self._encoding = launcher.encoding self._type += "; charset=" + launcher.encoding else: self._encoding = None def launch(self): self.logger.info("adding static route: %s", self._urlpath) self._app.route(self._urlpath, 'GET', self) def _parse_date(self, ims): """ Parse rfc1123, rfc850 and asctime timestamps and return UTC epoch. """ try: ts = email.utils.parsedate_tz(ims) return time.mktime(ts[:8] + (0,)) - (ts[9] or 0) - time.timezone except (TypeError, ValueError, IndexError, OverflowError): return None def __call__(self): # Get file contents and time stamp. If we can't, return an # appropriate HTTP error response. try: with open(self._fspath, "rb") as fp: mtime = os.fstat(fp.fileno()).st_mtime bytes = fp.read() except FileNotFoundError as e: return bottle.HTTPError(status=404, exception=e) except PermissionError as e: return bottle.HTTPError(status=403, exception=e) except OSError as e: self.logger.exception("unexpected exception reading %r", self._fspath) return bottle.HTTPError(status=500, exception=e) # Establish preliminary standard headers. headers = { "Content-Type": self._type, "Last-Modified": time.strftime("%a, %d %b %Y %H:%M:%S GMT", time.gmtime(mtime)), } if self._encoding: headers["Content-Encoding"] = self._encoding # Support the If-Modified-Since request header. ims = bottle.request.environ.get('HTTP_IF_MODIFIED_SINCE') if ims: ims = self._parse_date(ims.split(";")[0].strip()) if ims is not None and ims >= int(mtime): headers["Content-Length"] = "0" return bottle.HTTPResponse(body=b"", status=304, headers=headers) # Standard response. headers["Content-Length"] = str(len(bytes)) return bottle.HTTPResponse(body=bytes, status=200, headers=headers) class _TinCanErrorRoute(object): """ A route to an error page. These don't get routes created for them, and are only reached if an error routes them there. Unless you create custom code-behind, only two variables are available to your template: request (bottle.Request) and error (bottle.HTTPError). """ def __init__(self, template, loads, klass, lock, logger): self._template = template self._template.prepare() self._loads = loads self._class = klass self.lock = lock self.logger = logger def __call__(self, e): bottle.request.environ[_FTYPE] = True try: obj = self._class(bottle.request, e) obj.handle() tvars = self._loads.copy() tvars.update(obj.export()) with self.lock: return self._template.render(tvars).lstrip('\n') except bottle.HTTPResponse as e: return e except Exception as e: self.logger.exception("unexpected exception in error page") # Bottle doesn't allow error handlers to themselves cause # errors, most likely as a measure to prevent looping. So # this will cause a "Critical error while processing request" # page to be displayed, and any installed error pages to be # ignored. raise bottle.HTTPError(status=500, exception=e) class _TinCanRoute(_TinCanBaseRoute): """ A route created by the TinCan launcher. """ def __init__(self, launcher, name, subdir): super().__init__(launcher, name, subdir) self._fsroot = launcher.fsroot self._urlroot = launcher.urlroot self._name = name self._python = name + _PEXTEN self._fspath = os.path.join(launcher.fsroot, *subdir, name + _TEXTEN) self._urlpath = self.urljoin(launcher.urlroot, *subdir, name + _TEXTEN) self._origin = self._urlpath self._subdir = subdir self._seen = set() self._app = launcher.app self._encoding = launcher.encoding def launch(self): """ Launch a single page. """ # Build master and header objects, process #forward directives oheader = None while True: try: self._template = TemplateFile(self._fspath) except IOError as e: if oheader is not None: note = "{0}: invalid #forward: ".format(self._origin) else: note = "" raise TinCanError("{0}{1!s}".format(note, e)) from e try: self._header = TemplateHeader(self._template.header) except TemplateHeaderError as e: raise TinCanError("{0}: {1!s}".format(self._fspath, e)) from e if oheader is None: oheader = self._header # save original header elif (oheader.errors is None) != (self._header.errors is None): raise TinCanError("{0}: invalid #forward".format(self._origin)) if self._header.forward is None: break # self.logger.debug("forwarding from: %s", self._urlpath) # debug self._redirect() # self.logger.debug("forwarded to: %s", self._urlpath) # debug # If this is a #hidden page, we ignore it for now, since hidden pages # don't get routes made for them. if oheader.hidden and not oheader.errors: return # Get the code-behind #python if self._header.python is None: self._python_specified = False else: if not self._header.python.endswith(_PEXTEN): raise TinCanError("{0}: #python files must end in {1}".format(self._urlpath, _PEXTEN)) self._python = self._header.python self._python_specified = True # Obtain a class object by importing and introspecting a module. self._getclass() # Build body object (#template) and obtain #loads. if self._header.template is not None: if not self._header.template.endswith(_TEXTEN): raise TinCanError("{0}: #template files must end in {1}".format(self._urlpath, _TEXTEN)) try: rtpath = self._splitpath(self._header.template) tpath = os.path.normpath(os.path.join(self._fsroot, *rtpath)) tfile = TemplateFile(tpath) except OSError as e: raise TinCanError("{0}: invalid #template: {1!s}".format(self._urlpath, e)) from e except IndexError as e: raise TinCanError("{0}: invalid #template".format(self._urlpath)) from e self._body = ChameleonTemplate(source=tfile.body, encoding=self._encoding) else: self._body = ChameleonTemplate(source=self._template.body, encoding=self._encoding) self._body.prepare() # Process loads self._loads = {} for load in self._header.load: try: load = _LoadedFile(load) except ValueError as e: raise TinCanError("{0}: bad #load: {1!s}".format(self._urlpath, e)) from e if load.in_lib: fdir = os.path.join(self._fsroot, _WINF, "tlib") else: fdir = os.path.join(self._fsroot, *self._subdir) try: tmpl = self.get_template(load.fname, fdir, self._encoding) except Exception as e: raise TinCanError("{0}: bad #load: {1!s}".format(self._urlpath, e)) from e self._loads[load.vname] = tmpl.tpl # If this is an #errors page, register it as such. if oheader.errors is not None: self._mkerror(oheader.errors) return # this implies #hidden # Get #methods for this route if self._header.methods is None: methods = [ 'GET' ] else: methods = [ i.upper() for i in self._header.methods.split() ] if not methods: raise TinCanError("{0}: no #methods specified".format(self._urlpath)) # Register this thing with Bottle self.logger.info("adding route: %s (%s)", self._origin, ','.join(methods)) self._app.route(self._origin, methods, self) def _splitpath(self, unsplit): return _normpath(self._subdir, unsplit) def _mkerror(self, rerrors): try: errors = [ int(i) for i in rerrors.split() ] except ValueError as e: raise TinCanError("{0}: bad #errors line".format(self._urlpath)) from e if not errors: errors = range(_ERRMIN, _ERRMAX+1) route = _TinCanErrorRoute( ChameleonTemplate(source=self._template.body, encoding=self._encoding), self._loads, self._class, self.lock, self.logger) for error in errors: if error < _ERRMIN or error > _ERRMAX: raise TinCanError("{0}: bad #errors code".format(self._urlpath)) self._app.error_handler[error] = route # XXX def _gettime(self, path): try: return os.stat(path).st_mtime except FileNotFoundError: return 0 except OSError as e: raise TinCanError(str(e)) from e def _getclass(self): try: pypath = os.path.normpath(os.path.join(self._fsroot, *self._splitpath(self._python))) except IndexError as e: raise TinCanError("{0}: invalid #python".format(self._urlpath)) from e klass = ErrorPage if self._header.errors else Page # Give 'em a default code-behind if they don't furnish one pytime = self._gettime(pypath) if not pytime: if self._python_specified: raise TinCanError("{0}: #python file not found".format(self._urlpath)) self._class = klass return # Else load the code-behind from a .py file pycpath = pypath + 'c' pyctime = self._gettime(pycpath) try: if pyctime < pytime: py_compile.compile(pypath, cfile=pycpath, doraise=True) except py_compile.PyCompileError as e: raise TinCanError(str(e)) from e except Exception as e: raise TinCanError("{0}: {1!s}".format(pypath, e)) from e try: spec = importlib.util.spec_from_file_location(_mangle(self._name), pycpath) mod = importlib.util.module_from_spec(spec) spec.loader.exec_module(mod) except Exception as e: raise TinCanError("{0}: error importing: {1!s}".format(pycpath, e)) from e # Locate a suitable class. We look for the "deepest" class object # we can find in the inheritance tree. self._class = None score = -1 ambig = False for i in dir(mod): v = getattr(mod, i) if not isclass(v): continue d = self._cldepth(klass, v) if d > score: self._class = v score = d ambig = False elif d == score: ambig = True if self._class is None: raise TinCanError("{0}: contains no {1} classes".format(pypath, klass.__name__)) if ambig: raise TinCanError("{0}: contains ambiguous {1} classes".format(pypath, klass.__name__)) # This might fail for complex inheritance schemes from the classes of # interest (so don't use them!). def _cldepth(self, base, klass, count=0): if klass is object: # not found return -1 elif klass is base: # just found return count else: # must recurse for c in klass.__bases__: result = self._cldepth(base, c, count=count+1) if result > 0: return result return -1 def _redirect(self): try: rlist = self._splitpath(self._header.forward) forw = '/' + '/'.join(rlist) if forw in self._seen: raise TinCanError("{0}: #forward loop".format(self._origin)) self._seen.add(forw) rname = rlist.pop() except IndexError as e: raise TinCanError("{0}: invalid #forward".format(self._urlpath)) from e name, ext = os.path.splitext(rname) if ext != _TEXTEN: raise TinCanError("{0}: invalid #forward".format(self._urlpath)) self._subdir = rlist self._python = name + _PEXTEN self._fspath = os.path.join(self._fsroot, *self._subdir, rname) self._urlpath = '/' + self.urljoin(*self._subdir, rname) def __call__(self): """ This gets called by the framework AFTER the page is launched. """ target = None try: obj = self._class(bottle.request, bottle.response) obj.handle() tvars = self._loads.copy() tvars.update(obj.export()) with self.lock: return self._body.render(tvars).lstrip('\n') except ForwardException as fwd: target = fwd.target except bottle.HTTPResponse as e: return e except Exception as e: self.logger.exception("%s: unexpected exception", self._urlpath) raise bottle.HTTPError(status=500, exception=e) if target is None: self.logger.error("%s: unexpected null target", self._urlpath) raise bottle.HTTPError(status=500, exception=TinCanError(message)) # We get here if we are doing a server-side programmatic # forward. environ = bottle.request.environ if _FORIG not in environ: environ[_FORIG] = self._urlpath if _FLOOP not in environ: environ[_FLOOP] = set([self._urlpath]) elif target in environ[_FLOOP]: self.logger.error("%s: forward loop detected", environ[_FORIG]) raise bottle.HTTPError(status=500, exception=TinCanError(message)) environ[_FLOOP].add(target) environ['bottle.raw_path'] = target environ['PATH_INFO'] = urllib.parse.quote(target) route, args = self._app.router.match(environ) environ['route.handle'] = environ['bottle.route'] = route environ['route.url_args'] = args return route.call(**args) # M u t e x # # A dummy lock class, which is what we use if we don't need locking. class _DummyLock(object): def acquire(self, blocking=True, timeout=-1): pass def release(self): pass def __enter__(self): self.acquire() return self def __exit__(self, exc_type, exc_value, traceback): self.release() return False # L a u n c h e r _WINF = "WEB-INF" _BANNED = set([_WINF]) _EBANNED = set([_IEXTEN, _TEXTEN, _PEXTEN, _PEXTEN+"c"]) ENCODING = "utf-8" _BITBUCKET = logging.getLogger(__name__) _BITBUCKET.addHandler(logging.NullHandler) class _Launcher(object): """ Helper class for launching webapps. """ def __init__(self, fsroot, urlroot, multithread): """ Lightweight constructor. The real action happens in .launch() below. """ self.fsroot = fsroot self.urlroot = urlroot self.app = None self.errors = 0 self.debug = False self.encoding = ENCODING self.static = False self.multithread = multithread self.logger = _BITBUCKET def launch(self): """ Does the actual work of launching something. XXX - modifies sys.path and never un-modifies it. """ # Sanity checks if not self.urlroot.startswith("/"): self.errors = 1 self.logger("urlroot not absolute: {0!r}".format(self.urlroot)) return self if not os.path.isdir(self.fsroot): self.errors = 1 self.logger("no such directory: {0!r}".format(self.fsroot)) return self # Make any needed directories. Refuse to launch things that don't # contain WEB-INF, to prevent accidental launches of undesired # directory trees containing sensitive files. winf = os.path.join(self.fsroot, _WINF) if not os.path.isdir(winf): self.errors = 1 self.logger("no WEB-INF directory in {0!r}".format(self.fsroot)) return self lib = os.path.join(winf, "lib") for i in [ lib ]: if not os.path.isdir(i): os.mkdir(i) # Add our private lib directory to sys.path sys.path.insert(1, os.path.abspath(lib)) # Do what we gotta do self.app = TinCan() self._launch([]) return self def _launch(self, subdir): for entry in os.listdir(os.path.join(self.fsroot, *subdir)): if entry.startswith("."): continue # hidden file if not subdir and entry in _BANNED: continue etype = os.stat(os.path.join(self.fsroot, *subdir, entry)).st_mode if S_ISREG(etype): ename, eext = os.path.splitext(entry) if eext == _TEXTEN: route = _TinCanRoute(self, ename, subdir) else: if eext in _EBANNED: continue if self.static: route = _TinCanStaticRoute(self, entry, subdir) else: continue try: route.launch() except TinCanError as e: if self.logger.getEffectiveLevel() <= logging.DEBUG: self.logger.exception(str(e)) else: self.logger.error(str(e)) self.errors += 1 elif S_ISDIR(etype): self._launch(subdir + [entry]) def launch(fsroot=None, urlroot='/', multithread=True, **kwargs): """ Launch and return a TinCan webapp. Does not run the app; it is the caller's responsibility to call app.run() """ if fsroot is None: fsroot = os.getcwd() launcher = _Launcher(fsroot, urlroot, multithread) for i in [ "logger", "encoding", "static" ]: if i in kwargs: setattr(launcher, i, kwargs[i]) launcher.launch() return launcher.app, launcher.errors # XXX - We cannot implement a command-line launcher here; see the # launcher script for why.