curl tool globbing range [1-1] doesn't work

[rdsteed]

I recently upgraded curl.

My old version allowed allowed the curl command to use single value ranges eg [1-1]. In the new version this causes an error.

Is there any chance that curl could revert to the older, more permissive behavior?

[jay]

Please fill out the template, it is not filled out

[rdsteed]

Sorry about the initial post missing information - snafu with github editor.

Addendum - current version information:
curl 7.52.1 (i386-pc-win32) libcurl/7.52.1 OpenSSL/1.0.2k zlib/1.2.8 nghttp2/1.19.0

Old version is unknown - lost in hard disk crash. Downloaded latest curl as part of recovery.

[jay]

[1-1] was supported starting in b169aa2 but it likely broke in 5ca96cb which has glob fail when this is true: (step_n > (max_n - min_n)) or (1 > (1 - 1)). There is similar logic to stop something like [a-a]. I think that could be supported again, unless it was done intentionally. @bagder?

diff --git a/src/tool_urlglob.c b/src/tool_urlglob.c
index 0edfac6..b0e46b2 100644
--- a/src/tool_urlglob.c
+++ b/src/tool_urlglob.c
@@ -220,8 +220,9 @@ static CURLcode glob_range(URLGlob *glob, char **patternp,
 
     *posp += (pattern - *patternp);
 
-    if((rc != 3) || (min_c >= max_c) || ((max_c - min_c) > ('z' - 'a')) ||
-       (step <= 0) )
+    if(rc != 3 || step <= 0 ||
+       (min_c == max_c && step != 1) ||
+       (min_c != max_c && (min_c > max_c || (max_c - min_c) > ('z' - 'a'))))
       /* the pattern is not well-formed */
       return GLOBERROR("bad range", *posp, CURLE_URL_MALFORMAT);
 
@@ -230,9 +231,9 @@ static CURLcode glob_range(URLGlob *glob, char **patternp,
     pat->content.CharRange.ptr_c = pat->content.CharRange.min_c = min_c;
     pat->content.CharRange.max_c = max_c;
 
-    if(multiply(amount, (pat->content.CharRange.max_c -
+    if(multiply(amount, ((pat->content.CharRange.max_c -
                           pat->content.CharRange.min_c) /
-                         pat->content.CharRange.step + 1) )
+                         pat->content.CharRange.step + 1)))
       return GLOBERROR("range overflow", *posp, CURLE_URL_MALFORMAT);
   }
   else if(ISDIGIT(*pattern)) {
@@ -293,7 +294,9 @@ static CURLcode glob_range(URLGlob *glob, char **patternp,
     fail:
     *posp += (pattern - *patternp);
 
-    if(!endp || (min_n > max_n) || (step_n > (max_n - min_n)) || !step_n)
+    if(!endp || !step_n ||
+       (min_n == max_n && step_n != 1) ||
+       (min_n != max_n && (min_n > max_n || step_n > (max_n - min_n))))
       /* the pattern is not well-formed */
       return GLOBERROR("bad range", *posp, CURLE_URL_MALFORMAT);
 
@@ -303,9 +306,9 @@ static CURLcode glob_range(URLGlob *glob, char **patternp,
     pat->content.NumRange.max_n = max_n;
     pat->content.NumRange.step = step_n;
 
-    if(multiply(amount, (pat->content.NumRange.max_n -
-                         pat->content.NumRange.min_n) /
-                        pat->content.NumRange.step + 1) )
+    if(multiply(amount, ((pat->content.NumRange.max_n -
+                          pat->content.NumRange.min_n) /
+                         pat->content.NumRange.step + 1)))
       return GLOBERROR("range overflow", *posp, CURLE_URL_MALFORMAT);
   }
   else

[bagder]

I made them not work (at some point when I went through and tightened up the range handling), because 1-1 isn't really range, it is just a single value and thus I think it is more likely to hint that there was a mistake made than that the user actually asks for a single value written in a clumsy way...

[jay]

You may have a script that sets the range for a series of transfers, and maybe there's only one transfer to make. @rdsteed what's your use case?

[rdsteed]

@jay that is exactly the case. It is from a Windows batch file wrapping a curl command for downloading of podcast files. 1 to many years, 1 to many months, 1 to many days in one command. It also facilitated the #n notation in curl output for naming the downloads.

@Badger I found the [1-1] consistent with being able to use (a1..a1) notation in Excel spreadsheets to specify a single cell range or [1:2] to specify a single list item in Python. Of course, it doesn't make a whole lot of sense in these examples where constants are used, but it actually makes a lot of sense when the curl command is in a batch file and there is substitution of command arguments.

[bagder]

ok, so let's fix this and add a test or two to make sure it sticks

[rdsteed]

Thanks.
I'll just add that my expectation was that curl was interpreting the bracket notation to a for loop equivalent as follows:

[start-stop:step]
becomes
for (i = start; i <= stop; i+=step)

With this in mind
start > stop results in a zero trip loop
start=stop results in a one trip loop
step > (stop-start) results in a one trip loop.

I don't believe one trip loops should cause an error, but a warning in verbose mode could be helpful.

Finally, though I have a reading knowledge of C, I've never written in it. Otherwise, I would have contributed a pull request or a patch. Thanks for doing this.

[rdsteed]

Oops. Didn't mean to close. Clicked the wrong button. Still not used to this new touchpad.

[bagder]

@jay, will you bring this forward? You can probably make a test for it based on test87. If not, let me know and I can work on it.

[jay]

Ok, I added a test and made a branch for review. It is slightly different from my patch above because I had to make some adjustments to allow for [a-a:1].
https://github.com/curl/curl/compare/master...jay:fix_glob_range?expand=1

[bagder]

Looks good to me!

[jay]

Thanks and thanks for your report @rdsteed, fix just landed in 7a9f574.