[Therion] Filename resolution for source and input is inconsistent with documented Therion syntax

[mike at mikemccombe.co.uk]

In the description of the "input" statement, The Therion Book states:

"Default extension is .th and may be omitted."

I understand this to mean that:

"input bigcave" is equivalent to: "input bigcave.th" rather than meaning
that an actual extensionless file named "bigcave" is also a valid Therion
source.

 

Both "input" and "source" ultimately use thinput to open files. The
filename-resolution logic appears to try an extensionless pathname before
applying the configured extensions. Thus, given:

"input bigcave"

Therion effectively tries:

bigcave
bigcave.th
bigcave.th2

stopping at the first pathname it can open.

This causes a platform-dependent failure if a directory called "bigcave"
exists alongside "bigcave.th".

Under Windows, attempting to open the directory as an input stream fails, so
Therion continues and successfully opens "bigcave.th".

Under Linux, opening the directory as an input stream succeeds. Therion
therefore accepts the directory as its source and subsequently attempts to
read it as a text file. The stream enters a failure state, which Therion
reports with the misleading fatal error:

"Line too long"

The same problem occurs with "source".

 

Minimal reproduction:

Create a directory containing:

thconfig
bigcave.th
bigcave/ (an empty directory)

with thconfig containing:

"source bigcave"

and bigcave.th containing valid Therion source.

 

On Linux this fails with "line too long". Any of the following makes it
work:

*	Change the configuration to "source bigcave.th". 
*	Remove or rename the "bigcave" directory. 
*	Compile the same ambiguous project under Windows. 

This suggests that the underlying problem is that "the .th extension may be
omitted" has been implemented as "first try opening the extensionless
pathname" rather than "if the extension is omitted, use the default .th
extension."

There is a related issue with using the same generic filename-resolution
mechanism for both "source" and "input". I believe the intended semantics
are:

 

input bigcave -> bigcave.th
input bigcave.th -> bigcave.th
input bigcave.th2 -> bigcave.th2

 

source bigcave -> bigcave.th
source bigcave.th -> bigcave.th
source bigcave.th2 -> invalid

 

An explicitly specified .th2 file is therefore a valid target for "input",
but that does not make .th2 an alternative default extension when the
extension is omitted. Similarly, an actual extensionless filesystem object
is not a valid interpretation of either statement.

In other words, thinput appears to test filesystem objects that are not
valid interpretations of the Therion statement that invoked it.

Resolving the filename according to the semantics of "source" or "input"
before attempting to open it would also remove the Windows/Linux difference
and prevent a directory from being mistaken for a Therion source file.

 

Regards,

Mike

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.speleo.sk/pipermail/therion/attachments/20261004/39d84f45/attachment.htm>