source: main/trunk/greenstone2/build-src/packages/wget/README@ 31829

wget - web site mirroring software

This directory contains a version of the GNU wget program that was written by
Hrvoje Niksic and others and distributed under the GNU General Public Licence

The GNU wget homepage is http://www.cg.tuwien.ac.at/~prikryl/wget.html

1) A couple of very small changes were made to this package by Stefan Boddie
([email protected]). The USE_STDARG preprocessor definition was
added to the VC++ makefile as it appeared to need it to compile on either
VC++ 4.2 or VC++ 6.0. A change was also made to prevent backslashes from
mistakenly being converted to "@5c" on windows.

2) 2003/11/27 - [email protected] made a few more small changes to fix
h_errno bug preventing code from compiling on latest versions of GCC.

3) 2004/06/30 - [email protected] replaced version 1.5.3 with version 1.9.
It compiled as is under Windows XP, using VC++ 6.0, under Linux using GCC 3.2.3
and GCC 2.95.3. So I haven't made any changes. 
Under Windows, I haven't compiled it with openssl, so it doesn't support
https://. OpenSSL comes with a warning about it being illegal in some countries
to distribute cryptographic technology, so I haven't used it. Look at
wget-1.9/windows/README for Windows compiling instructions if you want to add
it in.

4) 2008/07/03 - [email protected] replaced version 1.9 with version 1.11.4.
Version 1.9 didn't compile statically with the new libraries which come with
fedora, so upgrading. Had to apply the patch at
http://marc.info/?l=wget&m=115131792408685&w=2 to get it to compile under
windows(VC++ 6.0), and those changes are in the tar.gz file in the repository.
Again, no ssl in windows wget. The packages/configure and packages/Makefile
files perform the same operations on wget as before, just the directory name
has changed from wget/wget-1.9 to wget/wget-1.11.4.


5) 2013/01/09 - [email protected] Make minor change to src/Makefile.in so the code compiles under minGW cross-compiler running on Linux
diff -r wget-1.13.4/src/Makefile.in wget-1.13.4.MINGW/src/Makefile.in
1136a1137,1143
> 
> ifeq ($(GSDLOS),windows)
>   # Compiling wget with MinGW
>   CFLAGS += -DIN6_ARE_ADDR_EQUAL=IN6_ADDR_EQUAL
>   LIBS += -lws2_32
> endif
> 

These newly added lines should come after the line:
> CLEANFILES = *~ *.bak core core.[0-9]* build_info.c version.c

and before the command to make 'all':
> all: config.h
>   $(MAKE) $(AM_MAKEFLAGS) all-am


6) 2013/01/29 - [email protected] added some #ifdef __ANDROID__ blocks into src/util.c to cope with the fact that this OS lacks localeconv()

diff wget-1.13.4-orig/src/utils.c wget-1.13.4-gs/src/utils.c

1466a1467,1470
> #ifdef __ANDROID__
>       cached_sep =",";
>       cached_grouping = "\x03";
> #else
1489a1494
> #endif

[Now the #endif comes around line 1441 or 1459, immediately before:
> initialized = true;
>    }
>  *sep = cached_sep;
>  *grouping = cached_grouping;]


7) 2014/10/13 - ak19
Moved to wget version 1.15. 
Only the changes numbered 5 and 6 above have been ported into it, following Dr Bainbridge's instructions, as they were both changes made to the previous wget version Greenstone used (1.13.4).
The wget tar file name (wget-1.15-gs) now indicates the version number and that it has been modified for Greenstone, so the Makefile and configure file in build-src/packages/ have been updated to reflect this.

8) 2017/07/27 - ak19 ([email protected]) - still using wget version 1.15, but now compiling wget up with OpenSSL support. Wget needs SSL support in order for it to access pages over HTTPS. In future, the web will be using https.

We're now compiling up OpenSSL during the configuration phase since wget needs it to exist during its configure phase. We;re building OpenSSL statically, by setting the no-shared flag. The built OpenSSL gets put into gs2build/linux|darwin/openssl, containing lib, include and bin subfolders. When configuring wget, we build wget against our OpenSSl, and make and make install proceed as normal. Refer to gs2build/build-src/packages configure.

We weren't compiling up wget statically before either, so we're still not doing so. To compile up wget (statically or not) with openssl, a helpful page was
https://stackoverflow.com/questions/9817337/compiling-wget-with-static-linking-self-compiled-openssl-library-linking-issu
Note, however, that since the CPPFLAGS and LDFLAGS are now set to point to our OpenSSL during the configure stage, the make command needn't additionally set them as well, contrary to the instruction for make on the stackoverflow page. So we just need to do the usual make, make install once the configure is done against OpenSSL.

If compiling wget up statically, then, in the LDFLAGS prepended to wget's configure command, append -static. Further, the gcc command that gets run needs to have -lpthread in its library listing at the end. The order of the libraries listed also needs to change for static compilation to be successful:
-lprce -lpthread -ldl <remaining -llibs>

However, warnings appear when compiling wget statically, as it does not make sense to create some programs statically since they may be stuck including a local context (e.g. something related to DNS warnings in compiling up a previous component statically). Linking against some libraries to create a static binary may not make sense either. For instance -ldl, the dynamic loading or linking library, may not make sense if the binary created is static. This seems to imply that wget makes more sense if compiled up as a shared object, .so, than as a static one, .a.

The existing version of wget, 1.15, works with HTTPS when compiled against OpenSSL. However, this version of the binary needs to be run with the --no-check-certificate flag on to access https pages without a security certificate.

e.g. ./wget --no-check-certificate http://englishhistory.net/tudor/citizens/

The system wget on Ubuntu 16.04 is version 1.17.1 and does not require this flag. Pre-compiled windows binaries are available for version 1.11.4, so that may still require the flag. This will require further investigation. We'd like both unix and windows operating systems to behave similarly, ideally.

* http://nebm.ist.utl.pt/~glopes/wget/
Prebuilt Windows wget binaries (for 32 and 64 bit) version 1.11.4 that includes SSL support
* http://gnuwin32.sourceforge.net/packages/wget.htm
GNU's prebuilt Windows binaries of wget v 1.11.4. May not have been built with SSL support.

The wget 1.11.4 we have, compiled without SSL does not work on https pages.
Neither does the wget from http://nebm.ist.utl.pt/~glopes/wget/, even after including the --no-check-certificate flag in the wget command. So we'll need to upgrade the wget binary on Windows to a later version too.


9) We're now shifting to wget-1.17.1 which is installed on Ubuntu 16.04, and for which a windows binary compiled with OpenSSL is available. Both the linux system version and windows binary work on https urls without the --no-check-certificate flag being necessary. However, the compiled up Linux version still needs this flag, see under PROBLEM.

This way our perl code can launch wget as before, without always passing that additional flag. Hopefully the output in the Download pane will be the same so that the donwload parsing will work.

LINUX CHANGES:
- Grabbed wget-1.17.1.tar.gz from https://ftp.gnu.org/gnu/wget/
- Made the modifications in steps 5 and 6 above and re-tarred as wget-1.17.1-gs.tar.gz, with corresponding changes in build-src/packages' configure, Makefile.in and Makefile
- The configure step has now changed for wget v 1.17.1. Refer to build-src/packages/configure

The configure step requires setting
* --with-libssl-prefix to $bindir/openssl, so the wget build process can find openssl's include and lib folders. (Whereas the --with-ssl indicates what type of ssl we're using, which is openssl in our case.)
* configuring had initially failed, reporting that OPENSSL_CFLAGS and OPENSSL_LIBS need to be set if not wanting to use whatever pkg-config may find. To set LIBS variables, use one of these forms: LIBS="-L/path/to/lib" or LIBS="/path/to/lib/lib.a" or LIBS="-lssl". To combine all three, separate with spaces. See http://trac.greenstone.org/changeset/30948 and https://github.com/tatsuhiro-t/spdylay/issues/43

PROBLEM AND SOLUTION WITH WGETRC
Can turn off requiring a certificate check for https URLs in wgetrc conf file, as explained here:
https://superuser.com/questions/508696/wget-without-no-check-certificate
And the location of wgetrc: https://www.gnu.org/software/wget/manual/html_node/Wgetrc-Location.html (for now setting the env var WGETRC in $GSDLHOME/setup.bash, which points to $GSDLHOME/bin/linux). Then the locally built linux wget 1.17.1 works, and so would v 1.15.

However, the linux system wget and windows wget binary v 1.17.1 are not setting thisvariable in their wgetrc file, so how is the certificate check off for them by default? The windows wget binary v 1.15 does require the no-check-certificate flag, but v 1.17.1 works out of the box. So why does the 1.17.1 version I built on linux behave like the 1.15 on Windows binary (and built on Linux), which required the flag?


WINDOWS
Windows binaries for wget 1.7.11 and other versions, built with openSSL support, are at:
https://eternallybored.org/misc/wget/

I downloaded the 32 bit version "wget-1.17.1-win32.zip" from there (at https://eternallybored.org/misc/wget/releases/old/wget-1.17.1-win32.zip)

Unzipping wouldn't succeed, nor copying the zip's wget.exe directly, both producing a windows error message.
To successfully extract: use 7zip to view the contents of the zip, then rename the wget.exe to wget.not, then copy out the wget.not file and rename it back.

This version of wget on Windows 64 bit worked successfully to retrieve the https page of the Tudors site, and without the --no-check-certificate flag.

# http://osxdaily.com/2012/05/22/install-wget-mac-os-x/
# https://lists.gnu.org/archive/html/bug-wget/2014-12/msg00104.html

Alternatives for Windows:
Source:
- https://soliloquyforthefallen.net/?p=238
- https://github.com/wertarbyte/wget/tree/master/windows (README at end)
Binaries:
- https://stackoverflow.com/questions/14344921/wget-for-windows-7-trusted-source

COMBINING GREENSTONE's GPL with OpenSSL LICENSES
OpenSSL is under a double license, see https://www.openssl.org/source/license.html
The licenses for GPL and OpenSSL are incompatible, see https://www.gnu.org/licenses/license-list.en.html#OpenSSL
but you can combine it this way: https://opensource.stackexchange.com/questions/2233/gpl-v3-with-openssl-exception?rq=1
which is what we've done for GS2 and GS3.


TO DO:
- If I delete the gs2build/bin/linux/openssl folder, the built wget still works fine without it. Dr Bainbridge confirmed that this is because, wget is built against OpenSSL's static libraries and therefore no longer needs the OpenSSL stuff we build and have been putting into gs2build/bin/linux/openssl. So we no longer need to put the built OpenSSL there.

- Add a tick box in GLI > File > Preferences for turning on No Check Certificate over https, this should then replace our wgetrc file and env variable set in GS2's setup.bash. By default leave this flag unticked, so downloading won't work over https. Need to store this user setting in GLI's config.xml. Ensure that when the download over https failed, it results in an error.

- If the downloading error count > 0:
At the bottom of GLI > Download Pane > View Log > download error log - when we get errors:
You have the option of adjusting your proxy server settings (go through the Configure Proxy button)
For https certificate authentication, you have the option of turning off checking the certificate in the Connections tab of File > Preferences

Check the warnings on windows. If it's no longer always warning, then do the stuff above on warning too, not just on error.