一个无法转的CAJ格式样本，PyPDF2似乎有bug

[jsbyysheng]

PyPDF2似乎有bug，样本(度盘)及提取码: qvbm

[jsbyysheng]

I am sure this problem is associated with PyPDF2. I have tried the c rewrite version produced by @austerios in #45 and it worked well.

[HinTak]

@jsbyysheng Cannot download the caj file (please use the attach feature of the issue tracker, zip it up if needed), and the code for the c-rewrite is gone too - anybody has a clone to push it? Or just zip it up and put it here too?

[jsbyysheng]

@austerios's repo was in gitlab before and currently in here. Unfortunately, this repo cannot be compiled in the newest macOS directly. Minor changes should be applied. My caj sample is here.

[HinTak]

The software builds on linux but crashes on trying to run. caj2pdf seems to mostly got it done except adding the outline. Anyway, it is probably simple enough for somebody else to fix (sort of busy on jbig if I am spending time on caj2pdf at all...)

[HinTak]

@austerios the current code still crashes, and there are a few other portability problems too. Anyway, here is the portability diff ( my compiler no longer take -flto=thin, and -lconv is only needed on MacOS X or BSD variants, and not needed nor provided on linux at all). The rest is better debugging (-g for finding wherethe exact location of crash is, with a debugger) and removing optimization that's really not needed (this code doesn't need to be any faster, really...).

diff --git a/src/Makefile b/src/Makefile
index f8c1e34..f96aebd 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -15,8 +15,8 @@ obj = ${src:.c=.o}
 
 PREFIX = /usr/local
 
-CFLAGS = -O3 -march=native -pipe -flto=thin -Wall
-LDFLAGS = -Wl,-O3 -lcrypto -liconv -lz -Wl,--as-needed
+CFLAGS = -g -O2 -pipe -Wall
+LDFLAGS = -lcrypto -lz
 
 CFLAGS += -I/usr/local/include
 LDFLAGS += -L/usr/local/lib

here is the crash work-around ( I call it a work-around, as the proper fix is probably finding out why it is pointing at wrong memory there).

diff --git a/src/pdf.c b/src/pdf.c
index 202b6d2..7f5c9d5 100644
--- a/src/pdf.c
+++ b/src/pdf.c
@@ -58,7 +58,7 @@ pdf_obj_destroy(pdf_object_t **pdf)
                *pdf = (*pdf)->next;
                free(ptr->object);
                free(ptr->dictionary);
-               free(ptr->stream);
+               //free(ptr->stream);
                free(ptr);
        }
 }

I am posting these just in case he is watching this space (and he seems to be quite active on his repo even as recent as 8 days ago).

I have a converted outcome from melon now, so I can compare that with caj2pdf and see what works.

[jsbyysheng]

I am sorry that I cannot help you find out the problem because I am busy with my research recently. Below is the command I used to compile melon successfully in macOS. Wish it can help you!

cc -L/usr/local/opt/openssl/lib -L/usr/local/opt/libiconv/lib -I/usr/local/opt/openssl/include -I/usr/local/opt/libiconv/include -lcrypto -liconv -o melon melon.c cnki.c cnki_caj.c cnki_kdh.c cnki_nh.c cnki_outline_tree.c cnki_pdf.c cnki_xml.c pdf.c pdf_cnki.c pdf_get.c pdf_parser.c pdf_writer.c iconv.c

[HinTak]

That's fine @jsbyysheng - the message was mostly for https://git.noema.org/melon/ . So you are a mac os x user too?

Openssl on linux is at the system location, and iconv's functionality is built-in to the system (GNU) C library. So on linux, none of the Openssl switches, nor the iconv switches, are needed.

[HinTak]

Still a problem with current.