Re: [PATCH v3 5/7] convert: auto_crlf=false and no attributes set: same as binary

On 02/08/2016 07:27 PM, Junio C Hamano wrote:

tboegi@web.de writes:

quoted

From: Torsten Bögershausen <redacted>

When core.autocrlf is set to false, and no attributes are set, the file
is treated as binary.

This, and also on the title, I know by "binary" you mean "no
conversion is attempted", and it is the word used in the code around
there, but it still makes my heart skip a beat every time I read
this sentence--it is not like we do not treat the contents as text
after all.

In any case, I take the above sentence the statement of the fact,
describing how the world currently is, not declaring a new world
order.

The word binary is indeed not ideal here: no eol conversion is done.
(And the commit message could use "-text")

quoted

Simplify the logic and remove duplicated code when dealing with
(crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE) by
setting crlf_action=CRLF_BINARY already in convert_attrs().

I looked at all the places where CRLF_BINARY is checked.  The ones
that are in this patch are clearly where "Is it BINARY?" and "Is
AUTO_CRLF_FALSE and CRLF_GUESS both true?" mean the same thing, so
this is a correct simplification to these places.

It is not easy to see what the effect of this change to the other
places that use CRLF_BINARY, though.

 * output_eol() used to return EOL_UNSET when auto_crlf is not in
   effect and CRLF_GUESS is.  The function will see CRLF_BINARY with
   this patch in such a case, and returns EOL_UNSET.  So there is no
   change to the function and its callers.

OK

 * convert_attrs() has "If BINARY don't do anything and return".
   Will the patch change behaviour for the "not-autocrlf,
   CRLF_GUESS" case in this codepath?  I think ca->crlf_action used
   to be left as CRLF_GUESS here before the patch, and now by the
   time the control flow reaches here it is already CRLF_BINARY.
   Would it affect the callers, and if so how?

Not sure if I fully understand the question:

The old CRLF_GUESS could mean (a) core.autocrlf=true,
(b) core.autocrlf=input or (c) core.autocrlf=false.
The callers had to look at the core.autocrlf them self.
This patch removes (c), the next (or over next) (a) and (b)

    if (ca->crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE)
        ca->crlf_action = CRLF_BINARY;
}
The next patch 6/7 removes "GUESS" completely.

 * get_convert_attr_ascii() would change the behaviour, right?  It
   runs convert_attrs(), and with this change a path without
   attribute when autocrlf is not in effect would get BINARY and
   would show "-text", while the code before this change would give
   an empty string.  Am I misreading the code, or is the change
   intended?

Thanks.

(That happened in my first (not published) version,
today we have this:)
struct conv_attrs {
    struct convert_driver *drv;
    enum crlf_action attr_action; /* What attr says */
    enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
    int ident;
};

The idea is (after this and the next commit 6/7) is
that once we had run  convert_attrs() there is a clear picture
what should be done with the eols, and it is stored in crlf_action.

No further check with core.autocrlf in any calling party should be needed.

In a future series it should be possible to set
*text=auto
*eol=clrf
to behave exactly as if core.autocrlf=true.
But that is a change in behavior, and needs a separate series.
This series should not change the behavior, so a critical review is appreciated.


Does this answers the questions ?
Is a local amend possible ?