summaryrefslogtreecommitdiffstats
path: root/tests/util/zstream.c
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-05-04 12:47:08 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-05-04 12:47:08 +0000
commit29b5ab554790bb57337a3b6ab9dcd963cf69d22e (patch)
treebe1456d2bc6c1fb078695fad7bc8f6b212062d3c /tests/util/zstream.c
parentInitial commit. (diff)
downloadlibgit2-29b5ab554790bb57337a3b6ab9dcd963cf69d22e.tar.xz
libgit2-29b5ab554790bb57337a3b6ab9dcd963cf69d22e.zip
Adding upstream version 1.7.2+ds.upstream/1.7.2+ds
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'tests/util/zstream.c')
-rw-r--r--tests/util/zstream.c167
1 files changed, 167 insertions, 0 deletions
diff --git a/tests/util/zstream.c b/tests/util/zstream.c
new file mode 100644
index 0000000..5c89895
--- /dev/null
+++ b/tests/util/zstream.c
@@ -0,0 +1,167 @@
+#include "clar_libgit2.h"
+#include "zstream.h"
+
+static const char *data = "This is a test test test of This is a test";
+
+#define INFLATE_EXTRA 2
+
+static void assert_zlib_equal_(
+ const void *expected, size_t e_len,
+ const void *compressed, size_t c_len,
+ const char *msg, const char *file, const char *func, int line)
+{
+ z_stream stream;
+ char *expanded = git__calloc(1, e_len + INFLATE_EXTRA);
+ cl_assert(expanded);
+
+ memset(&stream, 0, sizeof(stream));
+ stream.next_out = (Bytef *)expanded;
+ stream.avail_out = (uInt)(e_len + INFLATE_EXTRA);
+ stream.next_in = (Bytef *)compressed;
+ stream.avail_in = (uInt)c_len;
+
+ cl_assert(inflateInit(&stream) == Z_OK);
+ cl_assert(inflate(&stream, Z_FINISH));
+ inflateEnd(&stream);
+
+ clar__assert_equal(
+ file, func, line, msg, 1,
+ "%d", (int)stream.total_out, (int)e_len);
+ clar__assert_equal(
+ file, func, line, "Buffer len was not exact match", 1,
+ "%d", (int)stream.avail_out, (int)INFLATE_EXTRA);
+
+ clar__assert(
+ memcmp(expanded, expected, e_len) == 0,
+ file, func, line, "uncompressed data did not match", NULL, 1);
+
+ git__free(expanded);
+}
+
+#define assert_zlib_equal(E,EL,C,CL) \
+ assert_zlib_equal_(E, EL, C, CL, #EL " != " #CL, __FILE__, __func__, (int)__LINE__)
+
+void test_zstream__basic(void)
+{
+ git_zstream z = GIT_ZSTREAM_INIT;
+ char out[128];
+ size_t outlen = sizeof(out);
+
+ cl_git_pass(git_zstream_init(&z, GIT_ZSTREAM_DEFLATE));
+ cl_git_pass(git_zstream_set_input(&z, data, strlen(data) + 1));
+ cl_git_pass(git_zstream_get_output(out, &outlen, &z));
+ cl_assert(git_zstream_done(&z));
+ cl_assert(outlen > 0);
+ git_zstream_free(&z);
+
+ assert_zlib_equal(data, strlen(data) + 1, out, outlen);
+}
+
+void test_zstream__fails_on_trailing_garbage(void)
+{
+ git_str deflated = GIT_STR_INIT, inflated = GIT_STR_INIT;
+ char i = 0;
+
+ /* compress a simple string */
+ git_zstream_deflatebuf(&deflated, "foobar!!", 8);
+
+ /* append some garbage */
+ for (i = 0; i < 10; i++) {
+ git_str_putc(&deflated, i);
+ }
+
+ cl_git_fail(git_zstream_inflatebuf(&inflated, deflated.ptr, deflated.size));
+
+ git_str_dispose(&deflated);
+ git_str_dispose(&inflated);
+}
+
+void test_zstream__buffer(void)
+{
+ git_str out = GIT_STR_INIT;
+ cl_git_pass(git_zstream_deflatebuf(&out, data, strlen(data) + 1));
+ assert_zlib_equal(data, strlen(data) + 1, out.ptr, out.size);
+ git_str_dispose(&out);
+}
+
+#define BIG_STRING_PART "Big Data IS Big - Long Data IS Long - We need a buffer larger than 1024 x 1024 to make sure we trigger chunked compression - Big Big Data IS Bigger than Big - Long Long Data IS Longer than Long"
+
+static void compress_and_decompress_input_various_ways(git_str *input)
+{
+ git_str out1 = GIT_STR_INIT, out2 = GIT_STR_INIT;
+ git_str inflated = GIT_STR_INIT;
+ size_t i, fixed_size = max(input->size / 2, 256);
+ char *fixed = git__malloc(fixed_size);
+ cl_assert(fixed);
+
+ /* compress with deflatebuf */
+
+ cl_git_pass(git_zstream_deflatebuf(&out1, input->ptr, input->size));
+ assert_zlib_equal(input->ptr, input->size, out1.ptr, out1.size);
+
+ /* compress with various fixed size buffer (accumulating the output) */
+
+ for (i = 0; i < 3; ++i) {
+ git_zstream zs = GIT_ZSTREAM_INIT;
+ size_t use_fixed_size;
+
+ switch (i) {
+ case 0: use_fixed_size = 256; break;
+ case 1: use_fixed_size = fixed_size / 2; break;
+ case 2: use_fixed_size = fixed_size; break;
+ }
+ cl_assert(use_fixed_size <= fixed_size);
+
+ cl_git_pass(git_zstream_init(&zs, GIT_ZSTREAM_DEFLATE));
+ cl_git_pass(git_zstream_set_input(&zs, input->ptr, input->size));
+
+ while (!git_zstream_done(&zs)) {
+ size_t written = use_fixed_size;
+ cl_git_pass(git_zstream_get_output(fixed, &written, &zs));
+ cl_git_pass(git_str_put(&out2, fixed, written));
+ }
+
+ git_zstream_free(&zs);
+ assert_zlib_equal(input->ptr, input->size, out2.ptr, out2.size);
+
+ /* did both approaches give the same data? */
+ cl_assert_equal_sz(out1.size, out2.size);
+ cl_assert(!memcmp(out1.ptr, out2.ptr, out1.size));
+
+ git_str_dispose(&out2);
+ }
+
+ cl_git_pass(git_zstream_inflatebuf(&inflated, out1.ptr, out1.size));
+ cl_assert_equal_i(input->size, inflated.size);
+ cl_assert(memcmp(input->ptr, inflated.ptr, inflated.size) == 0);
+
+ git_str_dispose(&out1);
+ git_str_dispose(&inflated);
+ git__free(fixed);
+}
+
+void test_zstream__big_data(void)
+{
+ git_str in = GIT_STR_INIT;
+ size_t scan, target;
+
+ for (target = 1024; target <= 1024 * 1024 * 4; target *= 8) {
+
+ /* make a big string that's easy to compress */
+ git_str_clear(&in);
+ while (in.size < target)
+ cl_git_pass(
+ git_str_put(&in, BIG_STRING_PART, strlen(BIG_STRING_PART)));
+
+ compress_and_decompress_input_various_ways(&in);
+
+ /* make a big string that's hard to compress */
+ srand(0xabad1dea);
+ for (scan = 0; scan < in.size; ++scan)
+ in.ptr[scan] = (char)rand();
+
+ compress_and_decompress_input_various_ways(&in);
+ }
+
+ git_str_dispose(&in);
+}